experiment_data.py

# This code is part of Qiskit.
#
# (C) Copyright IBM 2021.
#
# This code is licensed under the Apache License, Version 2.0. You may
# obtain a copy of this license in the LICENSE.txt file in the root directory
# of this source tree or at http://www.apache.org/licenses/LICENSE-2.0.
#
# Any modifications or derivative works of this code must retain this
# copyright notice, and modified files need to carry a notice indicating
# that they have been altered from the originals.
"""
Experiment Data class
"""

from __future__ import annotations
import logging
import dataclasses
import re
from typing import Dict, Optional, List, Union, Any, Callable, Tuple, TYPE_CHECKING
from datetime import datetime, timezone
from concurrent import futures
from threading import Event
from functools import wraps, singledispatch
from collections import deque
import contextlib
import copy
import uuid
import enum
import time
import io
import sys
import json
import traceback
import numpy as np
import pandas as pd
from dateutil import tz
from matplotlib import pyplot
from matplotlib.figure import Figure as MatplotlibFigure
from qiskit.result import Result
from qiskit.providers.jobstatus import JobStatus, JOB_FINAL_STATES
from qiskit.exceptions import QiskitError
from qiskit.providers import Job, Backend, Provider
from qiskit.utils.deprecation import deprecate_arg

from qiskit_ibm_experiment import (
    IBMExperimentService,
    ExperimentData as ExperimentDataclass,
    AnalysisResultData as AnalysisResultDataclass,
    ResultQuality,
)
from qiskit_experiments.framework.json import ExperimentEncoder, ExperimentDecoder
from qiskit_experiments.database_service.utils import (
    qiskit_version,
    plot_to_svg_bytes,
    ThreadSafeOrderedDict,
    ThreadSafeList,
)
from qiskit_experiments.database_service.device_component import to_component, DeviceComponent
from qiskit_experiments.framework.analysis_result import AnalysisResult
from qiskit_experiments.framework.analysis_result_data import AnalysisResultData
from qiskit_experiments.framework.analysis_result_table import AnalysisResultTable
from qiskit_experiments.framework import BackendData
from qiskit_experiments.database_service.exceptions import (
    ExperimentDataError,
    ExperimentEntryNotFound,
    ExperimentDataSaveFailed,
)

if TYPE_CHECKING:
    # There is a cyclical dependency here, but the name needs to exist for
    # Sphinx on Python 3.9+ to link type hints correctly.  The gating on
    # `TYPE_CHECKING` means that the import will never be resolved by an actual
    # interpreter, only static analysis.
    from . import BaseExperiment

LOG = logging.getLogger(__name__)


def do_auto_save(func: Callable):
    """Decorate the input function to auto save data."""

    @wraps(func)
    def _wrapped(self, *args, **kwargs):
        return_val = func(self, *args, **kwargs)
        if self.auto_save:
            self.save_metadata()
        return return_val

    return _wrapped


def utc_to_local(utc_dt: datetime) -> datetime:
    """Convert input UTC timestamp to local timezone.

    Args:
        utc_dt: Input UTC timestamp.

    Returns:
        A ``datetime`` with the local timezone.
    """
    if utc_dt is None:
        return None
    local_dt = utc_dt.astimezone(tz.tzlocal())
    return local_dt


def local_to_utc(local_dt: datetime) -> datetime:
    """Convert input local timezone timestamp to UTC timezone.

    Args:
        local_dt: Input local timestamp.

    Returns:
        A ``datetime`` with the UTC timezone.
    """
    if local_dt is None:
        return None
    utc_dt = local_dt.astimezone(tz.UTC)
    return utc_dt


def parse_utc_datetime(dt_str: str) -> datetime:
    """Parses UTC datetime from a string"""
    if dt_str is None:
        return None

    db_datetime_format = "%Y-%m-%dT%H:%M:%S.%fZ"
    dt_utc = datetime.strptime(dt_str, db_datetime_format)
    dt_utc = dt_utc.replace(tzinfo=timezone.utc)
    return dt_utc


class FigureData:
    """Wrapper class for figures and figure metadata. The raw figure can be accessed with
    the ``figure`` attribute."""

    def __init__(self, figure, name=None, metadata=None):
        """Creates a new figure data object.

        Args:
            figure: the raw figure itself. Can be SVG or matplotlib.Figure.
            name: Optional, the name of the figure.
            metadata: Optional, any metadata to be stored with the figure.
        """
        self.figure = figure
        self._name = name
        self.metadata = metadata or {}

    # name is read only
    @property
    def name(self) -> str:
        """The name of the figure"""
        return self._name

    @property
    def metadata(self) -> dict:
        """The metadata dictionary stored with the figure"""
        return self._metadata

    @metadata.setter
    def metadata(self, new_metadata: dict):
        """Set the metadata to new value; must be a dictionary"""
        if not isinstance(new_metadata, dict):
            raise ValueError("figure metadata must be a dictionary")
        self._metadata = new_metadata

    def copy(self, new_name: Optional[str] = None):
        """Creates a copy of the figure data"""
        name = new_name or self.name
        return FigureData(figure=self.figure, name=name, metadata=copy.deepcopy(self.metadata))

    def __json_encode__(self) -> Dict[str, Any]:
        """Return the json representation of the figure data"""
        return {"figure": self.figure, "name": self.name, "metadata": self.metadata}

    @classmethod
    def __json_decode__(cls, args: Dict[str, Any]) -> "FigureData":
        """Initialize a figure data from the json representation"""
        return cls(**args)

    def _repr_png_(self):
        if isinstance(self.figure, MatplotlibFigure):
            b = io.BytesIO()
            self.figure.savefig(b, format="png", bbox_inches="tight")
            png = b.getvalue()
            return png
        else:
            return None

    def _repr_svg_(self):
        if isinstance(self.figure, str):
            return self.figure
        if isinstance(self.figure, bytes):
            return str(self.figure)
        return None


_FigureT = Union[str, bytes, MatplotlibFigure, FigureData]


class ExperimentData:
    """Experiment data container class.

    .. note::
        Saving experiment data to the cloud database is currently a limited access feature. You can
        check whether you have access by logging into the IBM Quantum interface
        and seeing if you can see the `database <https://quantum-computing.ibm.com/experiments>`__.

    This class handles the following:

    1. Storing the data related to an experiment: raw data, metadata, analysis results,
       and figures
    2. Managing jobs and adding data from jobs automatically
    3. Saving and loading data from the database service

    The field ``db_data`` is a dataclass (``ExperimentDataclass``) containing
    all the data that can be stored in the database and loaded from it, and
    as such is subject to strict conventions.

    Other data fields can be added and used freely, but they won't be saved
    to the database.

    """

    _metadata_version = 1
    _job_executor = futures.ThreadPoolExecutor()

    _json_encoder = ExperimentEncoder
    _json_decoder = ExperimentDecoder

    _metadata_filename = "metadata.json"
    _max_workers_cap = 10

    def __init__(
        self,
        experiment: Optional["BaseExperiment"] = None,
        backend: Optional[Backend] = None,
        service: Optional[IBMExperimentService] = None,
        provider: Optional[Provider] = None,
        parent_id: Optional[str] = None,
        job_ids: Optional[List[str]] = None,
        child_data: Optional[List[ExperimentData]] = None,
        verbose: Optional[bool] = True,
        db_data: Optional[ExperimentDataclass] = None,
        start_datetime: Optional[datetime] = None,
        **kwargs,
    ):
        """Initialize experiment data.

        Args:
            experiment: Experiment object that generated the data.
            backend: Backend the experiment runs on. This overrides the
                backend in the experiment object.
            service: The service that stores the experiment results to the database
            provider: The provider used for the experiments
                (can be used to automatically obtain the service)
            parent_id: ID of the parent experiment data
                in the setting of a composite experiment
            job_ids: IDs of jobs submitted for the experiment.
            child_data: List of child experiment data.
            verbose: Whether to print messages.
            db_data: A prepared ExperimentDataclass of the experiment info.
                This overrides other db parameters.
            start_datetime: The time when the experiment started running.
                If none, defaults to the current time.

        Additional info:
            In order to save the experiment data to the cloud service, the class
            needs access to the experiment service provider. It can be obtained
            via three different methods, given here by priority:

            1. Passing it directly via the ``service`` parameter.
            2. Implicitly obtaining it from the ``provider`` parameter.
            3. Implicitly obtaining it from the ``backend`` parameter, using that backend's provider.
        """
        if experiment is not None:
            backend = backend or experiment.backend
            experiment_type = experiment.experiment_type
        else:
            # Don't use None since the resultDB won't accept that
            experiment_type = ""
        if job_ids is None:
            job_ids = []

        self._experiment = experiment

        # data stored in the database
        metadata = {}
        if experiment is not None:
            metadata = copy.deepcopy(experiment._metadata())
        source = metadata.pop(
            "_source",
            {
                "class": f"{self.__class__.__module__}.{self.__class__.__name__}",
                "metadata_version": self.__class__._metadata_version,
                "qiskit_version": qiskit_version(),
            },
        )
        metadata["_source"] = source
        experiment_id = kwargs.get("experiment_id", str(uuid.uuid4()))
        if db_data is None:
            self._db_data = ExperimentDataclass(
                experiment_id=experiment_id,
                experiment_type=experiment_type,
                parent_id=parent_id,
                job_ids=job_ids,
                metadata=metadata,
            )
        else:
            self._db_data = db_data
        if self.start_datetime is None:
            if start_datetime is None:
                start_datetime = datetime.now()
            self.start_datetime = start_datetime
        for key, value in kwargs.items():
            if hasattr(self._db_data, key):
                setattr(self._db_data, key, value)
            else:
                LOG.warning("Key '%s' not stored in the database", key)

        # general data related
        self._backend = None
        if backend is not None:
            self._set_backend(backend, recursive=False)
        self.provider = provider
        if provider is None and backend is not None:
            self.provider = backend.provider
        self._service = service
        if self._service is None and self.provider is not None:
            self._service = self.get_service_from_provider(self.provider)
        if self._service is None and self.provider is None and self.backend is not None:
            self._service = self.get_service_from_backend(self.backend)
        self._auto_save = False
        self._created_in_db = False
        self._extra_data = kwargs
        self.verbose = verbose

        # job handling related
        self._jobs = ThreadSafeOrderedDict(job_ids)
        self._job_futures = ThreadSafeOrderedDict()
        self._running_time = None
        self._analysis_callbacks = ThreadSafeOrderedDict()
        self._analysis_futures = ThreadSafeOrderedDict()
        # Set 2 workers for analysis executor so there can be 1 actively running
        # future and one waiting "running" future. This is to allow the second
        # future to be cancelled without waiting for the actively running future
        # to finish first.
        self._analysis_executor = futures.ThreadPoolExecutor(max_workers=2)
        self._monitor_executor = futures.ThreadPoolExecutor()

        # data storage
        self._result_data = ThreadSafeList()
        self._figures = ThreadSafeOrderedDict(self._db_data.figure_names)
        self._analysis_results = AnalysisResultTable()

        self._deleted_figures = deque()
        self._deleted_analysis_results = deque()

        # Child related
        # Add component data and set parent ID to current container
        self._child_data = ThreadSafeOrderedDict()
        if child_data is not None:
            self._set_child_data(child_data)

    # Getters/setters for experiment metadata

    @property
    def experiment(self):
        """Return the experiment for this data.

        Returns:
            BaseExperiment: the experiment object.
        """
        return self._experiment

    @property
    def completion_times(self) -> Dict[str, datetime]:
        """Returns the completion times of the jobs."""
        job_times = {}
        for job_id, job in self._jobs.items():
            if job is not None and "COMPLETED" in job.time_per_step():
                job_times[job_id] = job.time_per_step().get("COMPLETED")

        return job_times

    @property
    def tags(self) -> List[str]:
        """Return tags assigned to this experiment data.

        Returns:
            A list of tags assigned to this experiment data.

        """
        return self._db_data.tags

    @tags.setter
    def tags(self, new_tags: List[str]) -> None:
        """Set tags for this experiment."""
        if not isinstance(new_tags, list):
            raise ExperimentDataError(f"The `tags` field of {type(self).__name__} must be a list.")
        self._db_data.tags = np.unique(new_tags).tolist()
        if self.auto_save:
            self.save_metadata()

    @property
    def metadata(self) -> Dict:
        """Return experiment metadata.

        Returns:
            Experiment metadata.
        """
        return self._db_data.metadata

    @property
    def creation_datetime(self) -> datetime:
        """Return the creation datetime of this experiment data.

        Returns:
            The timestamp when this experiment data was saved to the cloud service
            in the local timezone.

        """
        return self._db_data.creation_datetime

    @property
    def start_datetime(self) -> datetime:
        """Return the start datetime of this experiment data.

        Returns:
            The timestamp when this experiment began running in the local timezone.

        """
        return self._db_data.start_datetime

    @start_datetime.setter
    def start_datetime(self, new_start_datetime: datetime) -> None:
        self._db_data.start_datetime = new_start_datetime

    @property
    def updated_datetime(self) -> datetime:
        """Return the update datetime of this experiment data.

        Returns:
            The timestamp when this experiment data was last updated in the service
            in the local timezone.

        """
        return self._db_data.updated_datetime

    @property
    def running_time(self) -> datetime:
        """Return the running time of this experiment data.

        The running time is the time the latest successful job started running on
        the remote quantum machine. This can change as more jobs finish.

        """
        return self._running_time

    @property
    def end_datetime(self) -> datetime:
        """Return the end datetime of this experiment data.

        The end datetime is the time the latest job data was
        added without errors; this can change as more jobs finish.

        Returns:
            The timestamp when the last job of this experiment finished
            in the local timezone.

        """
        return self._db_data.end_datetime

    @end_datetime.setter
    def end_datetime(self, new_end_datetime: datetime) -> None:
        self._db_data.end_datetime = new_end_datetime

    @property
    def hub(self) -> str:
        """Return the hub of this experiment data.

        Returns:
            The hub of this experiment data.

        """
        return self._db_data.hub

    @property
    def group(self) -> str:
        """Return the group of this experiment data.

        Returns:
            The group of this experiment data.

        """
        return self._db_data.group

    @property
    def project(self) -> str:
        """Return the project of this experiment data.

        Returns:
            The project of this experiment data.

        """
        return self._db_data.project

    @property
    def experiment_id(self) -> str:
        """Return experiment ID

        Returns:
            Experiment ID.
        """

        return self._db_data.experiment_id

    @property
    def experiment_type(self) -> str:
        """Return experiment type

        Returns:
            Experiment type.
        """

        return self._db_data.experiment_type

    @experiment_type.setter
    def experiment_type(self, new_type: str) -> None:
        """Sets the parent id"""
        self._db_data.experiment_type = new_type

    @property
    def parent_id(self) -> str:
        """Return parent experiment ID

        Returns:
            Parent ID.
        """
        return self._db_data.parent_id

    @parent_id.setter
    def parent_id(self, new_id: str) -> None:
        """Sets the parent id"""
        self._db_data.parent_id = new_id

    @property
    def job_ids(self) -> List[str]:
        """Return experiment job IDs.

        Returns: IDs of jobs submitted for this experiment.
        """
        return self._db_data.job_ids

    @property
    def figure_names(self) -> List[str]:
        """Return names of the figures associated with this experiment.

        Returns:
            Names of figures associated with this experiment.
        """
        return self._db_data.figure_names

    @property
    def share_level(self) -> str:
        """Return the share level for this experiment

        Returns:
            Experiment share level.
        """
        return self._db_data.share_level

    @share_level.setter
    def share_level(self, new_level: str) -> None:
        """Set the experiment share level,
           to this experiment itself and its descendants.

        Args:
            new_level: New experiment share level. Valid share levels are provider-
                specified. For example, IBM Quantum experiment service allows
                "public", "hub", "group", "project", and "private".
        """
        self._db_data.share_level = new_level
        for data in self._child_data.values():
            original_auto_save = data.auto_save
            data.auto_save = False
            data.share_level = new_level
            data.auto_save = original_auto_save
        if self.auto_save:
            self.save_metadata()

    @property
    def notes(self) -> str:
        """Return experiment notes.

        Returns:
            Experiment notes.
        """
        return self._db_data.notes

    @notes.setter
    def notes(self, new_notes: str) -> None:
        """Update experiment notes.

        Args:
            new_notes: New experiment notes.
        """
        self._db_data.notes = new_notes
        if self.auto_save:
            self.save_metadata()

    @property
    def backend_name(self) -> str:
        """Return the backend's name"""
        return self._db_data.backend

    @property
    def backend(self) -> Backend:
        """Return backend.

        Returns:
            Backend.
        """
        return self._backend

    @backend.setter
    def backend(self, new_backend: Backend) -> None:
        """Update backend.

        Args:
            new_backend: New backend.
        """
        self._set_backend(new_backend)
        if self.auto_save:
            self.save_metadata()

    def _set_backend(self, new_backend: Backend, recursive: bool = True) -> None:
        """Set backend.
        Args:
            new_backend: New backend.
            recursive: should set the backend for children as well
        """
        # defined independently from the setter to enable setting without autosave

        self._backend = new_backend
        self._backend_data = BackendData(new_backend)
        self._db_data.backend = self._backend_data.name
        if self._db_data.backend is None:
            self._db_data.backend = str(new_backend)
        provider = self._backend_data.provider
        if provider is not None:
            self._set_hgp_from_provider(provider)
        if recursive:
            for data in self.child_data():
                data._set_backend(new_backend)

    def _set_hgp_from_provider(self, provider):
        try:
            # qiskit-ibmq-provider style
            if hasattr(provider, "credentials"):
                creds = provider.credentials
                self.hgp = f"{creds.hub}/{creds.group}/{creds.project}"
            # qiskit-ibm-provider style
            if hasattr(provider, "_hgps"):
                for hgp_string, hgp in provider._hgps.items():
                    if self.backend.name in hgp.backends:
                        self.hgp = hgp_string
                        break
        except (AttributeError, IndexError, QiskitError):
            return

    @property
    def hgp(self) -> str:
        """Returns Hub/Group/Project data as a formatted string"""
        return f"{self.hub}/{self.group}/{self.project}"

    @hgp.setter
    def hgp(self, new_hgp: str) -> None:
        """Sets the Hub/Group/Project data from a formatted string"""
        if re.match(r"[^/]*/[^/]*/[^/]*$", new_hgp) is None:
            raise QiskitError("hgp can be only given in a <hub>/<group>/<project> format")
        self._db_data.hub, self._db_data.group, self._db_data.project = new_hgp.split("/")

    def _clear_results(self):
        """Delete all currently stored analysis results and figures"""
        # Schedule existing analysis results for deletion next save call
        self._deleted_analysis_results.extend(list(self._analysis_results.result_ids()))
        self._analysis_results.clear()
        # Schedule existing figures for deletion next save call
        for key in self._figures.keys():
            self._deleted_figures.append(key)
        self._figures = ThreadSafeOrderedDict()

    @property
    def service(self) -> Optional[IBMExperimentService]:
        """Return the database service.

        Returns:
            Service that can be used to access this experiment in a database.
        """
        return self._service

    @service.setter
    def service(self, service: IBMExperimentService) -> None:
        """Set the service to be used for storing experiment data

        Args:
            service: Service to be used.

        Raises:
            ExperimentDataError: If an experiment service is already being used.
        """
        self._set_service(service)

    @property
    def provider(self) -> Optional[Provider]:
        """Return the backend provider.

        Returns:
            Provider that is used to obtain backends and job data.
        """
        return self._provider

    @provider.setter
    def provider(self, provider: Provider) -> None:
        """Set the provider to be used for obtaining job data

        Args:
            provider: Provider to be used.
        """
        self._provider = provider

    @property
    def auto_save(self) -> bool:
        """Return current auto-save option.

        Returns:
            Whether changes will be automatically saved.
        """
        return self._auto_save

    @auto_save.setter
    def auto_save(self, save_val: bool) -> None:
        """Set auto save preference.

        Args:
            save_val: Whether to do auto-save.
        """
        # children will be saved once we set auto_save for them
        if save_val is True:
            self.save(save_children=False)
        self._auto_save = save_val
        for data in self.child_data():
            data.auto_save = save_val

    @property
    def source(self) -> Dict:
        """Return the class name and version."""
        return self._db_data.metadata["_source"]

    # Data addition and deletion

    def add_data(
        self,
        data: Union[Result, List[Result], Dict, List[Dict]],
    ) -> None:
        """Add experiment data.

        Args:
            data: Experiment data to add. Several types are accepted for convenience:

                * Result: Add data from this ``Result`` object.
                * List[Result]: Add data from the ``Result`` objects.
                * Dict: Add this data.
                * List[Dict]: Add this list of data.

        Raises:
            TypeError: If the input data type is invalid.
        """
        if any(not future.done() for future in self._analysis_futures.values()):
            LOG.warning(
                "Not all analysis has finished running. Adding new data may "
                "create unexpected analysis results."
            )
        if not isinstance(data, list):
            data = [data]

        # Directly add non-job data
        with self._result_data.lock:
            for datum in data:
                if isinstance(datum, dict):
                    self._result_data.append(datum)
                elif isinstance(datum, Result):
                    self._add_result_data(datum)
                else:
                    raise TypeError(f"Invalid data type {type(datum)}.")

    def add_jobs(
        self,
        jobs: Union[Job, List[Job]],
        timeout: Optional[float] = None,
    ) -> None:
        """Add experiment data.

        Args:
            jobs: The Job or list of Jobs to add result data from.
            timeout: Optional, time in seconds to wait for all jobs to finish
                     before cancelling them.

        Raises:
            TypeError: If the input data type is invalid.

        .. note::
            If a timeout is specified the :meth:`cancel_jobs` method will be
            called after timing out to attempt to cancel any unfinished jobs.

            If you want to wait for jobs without cancelling, use the timeout
            kwarg of :meth:`block_for_results` instead.
        """
        if any(not future.done() for future in self._analysis_futures.values()):
            LOG.warning(
                "Not all analysis has finished running. Adding new jobs may "
                "create unexpected analysis results."
            )
        if isinstance(jobs, Job):
            jobs = [jobs]

        # Add futures for extracting finished job data
        timeout_ids = []
        for job in jobs:
            if self.backend is not None:
                backend_name = BackendData(self.backend).name
                job_backend_name = BackendData(job.backend()).name
                if self.backend and backend_name != job_backend_name:
                    LOG.warning(
                        "Adding a job from a backend (%s) that is different "
                        "than the current backend (%s). "
                        "The new backend will be used, but "
                        "service is not changed if one already exists.",
                        job.backend(),
                        self.backend,
                    )
            self.backend = job.backend()

            jid = job.job_id()
            if jid in self._jobs:
                LOG.warning(
                    "Skipping duplicate job, a job with this ID already exists [Job ID: %s]", jid
                )
            else:
                self.job_ids.append(jid)
                self._jobs[jid] = job
                if jid in self._job_futures:
                    LOG.warning("Job future has already been submitted [Job ID: %s]", jid)
                else:
                    self._add_job_future(job)
                    if timeout is not None:
                        timeout_ids.append(jid)

        # Add future for cancelling jobs that timeout
        if timeout_ids:
            self._job_executor.submit(self._timeout_running_jobs, timeout_ids, timeout)

        if self.auto_save:
            self.save_metadata()

    def _timeout_running_jobs(self, job_ids, timeout):
        """Function for cancelling jobs after timeout length.

        This function should be submitted to an executor to run as a future.

        Args:
            job_ids: the IDs of jobs to wait for.
            timeout: The total time to wait for all jobs before cancelling.
        """
        futs = [self._job_futures[jid] for jid in job_ids]
        waited = futures.wait(futs, timeout=timeout)

        # Try to cancel timed-out jobs
        if waited.not_done:
            LOG.debug("Cancelling running jobs that exceeded add_jobs timeout.")
            done_ids = {fut.result()[0] for fut in waited.done}
            notdone_ids = [jid for jid in job_ids if jid not in done_ids]
            self.cancel_jobs(notdone_ids)

    def _add_job_future(self, job):
        """Submit new _add_job_data job to executor"""
        jid = job.job_id()
        if jid in self._job_futures:
            LOG.warning("Job future has already been submitted [Job ID: %s]", jid)
        else:
            self._job_futures[jid] = self._job_executor.submit(self._add_job_data, job)

    def _add_job_data(
        self,
        job: Job,
    ) -> Tuple[str, bool]:
        """Wait for a job to finish and add job result data.

        Args:
            job: the Job to wait for and add data from.

        Returns:
            A tuple (str, bool) of the job id and bool of if the job data was added.

        Raises:
            Exception: If an error occurred when adding job data.
        """
        jid = job.job_id()
        try:
            job_result = job.result()
            try:
                self._running_time = job.time_per_step().get("running", None)
            except AttributeError:
                pass
            self._add_result_data(job_result, jid)
            LOG.debug("Job data added [Job ID: %s]", jid)
            # sets the endtime to be the time the last successful job was added
            self.end_datetime = datetime.now()
            return jid, True
        except Exception as ex:  # pylint: disable=broad-except
            # Handle cancelled jobs
            status = job.status()
            if status == JobStatus.CANCELLED:
                LOG.warning("Job was cancelled before completion [Job ID: %s]", jid)
                return jid, False
            if status == JobStatus.ERROR:
                LOG.error(
                    "Job data not added for errored job [Job ID: %s]\nError message: %s",
                    jid,
                    job.error_message(),
                )
                return jid, False
            LOG.warning("Adding data from job failed [Job ID: %s]", job.job_id())
            raise ex

    def add_analysis_callback(self, callback: Callable, **kwargs: Any):
        """Add analysis callback for running after experiment data jobs are finished.

        This method adds the `callback` function to a queue to be run
        asynchronously after completion of any running jobs, or immediately
        if no running jobs. If this method is called multiple times the
        callback functions will be executed in the order they were
        added.

        Args:
            callback: Callback function invoked when job finishes successfully.
                      The callback function will be called as
                      ``callback(expdata, **kwargs)`` where `expdata` is this
                      ``DbExperimentData`` object, and `kwargs` are any additional
                      keyword arguments passed to this method.
            **kwargs: Keyword arguments to be passed to the callback function.
        """
        with self._job_futures.lock and self._analysis_futures.lock:
            # Create callback dataclass
            cid = uuid.uuid4().hex
            self._analysis_callbacks[cid] = AnalysisCallback(
                name=callback.__name__,
                callback_id=cid,
            )

            # Futures to wait for
            futs = self._job_futures.values() + self._analysis_futures.values()
            wait_future = self._monitor_executor.submit(
                self._wait_for_futures, futs, name="jobs and analysis"
            )

            # Create a future to monitor event for calls to cancel_analysis
            def _monitor_cancel():
                self._analysis_callbacks[cid].event.wait()
                return False

            cancel_future = self._monitor_executor.submit(_monitor_cancel)

            # Add run analysis future
            self._analysis_futures[cid] = self._analysis_executor.submit(
                self._run_analysis_callback, cid, wait_future, cancel_future, callback, **kwargs
            )

    def _run_analysis_callback(
        self,
        callback_id: str,
        wait_future: futures.Future,
        cancel_future: futures.Future,
        callback: Callable,
        **kwargs,
    ):
        """Run an analysis callback after specified futures have finished."""
        if callback_id not in self._analysis_callbacks:
            raise ValueError(f"No analysis callback with id {callback_id}")

        # Monitor jobs and cancellation event to see if callback should be run
        # or cancelled
        # Future which returns if either all jobs finish, or cancel event is set
        waited = futures.wait([wait_future, cancel_future], return_when="FIRST_COMPLETED")
        cancel = not all(fut.result() for fut in waited.done)

        # Ensure monitor event is set so monitor future can terminate
        self._analysis_callbacks[callback_id].event.set()

        # If not ready cancel the callback before running
        if cancel:
            self._analysis_callbacks[callback_id].status = AnalysisStatus.CANCELLED
            LOG.info(
                "Cancelled analysis callback [Experiment ID: %s][Analysis Callback ID: %s]",
                self.experiment_id,
                callback_id,
            )
            return callback_id, False

        # Run callback function
        self._analysis_callbacks[callback_id].status = AnalysisStatus.RUNNING
        try:
            LOG.debug(
                "Running analysis callback '%s' [Experiment ID: %s][Analysis Callback ID: %s]",
                self._analysis_callbacks[callback_id].name,
                self.experiment_id,
                callback_id,
            )
            callback(self, **kwargs)
            self._analysis_callbacks[callback_id].status = AnalysisStatus.DONE
            LOG.debug(
                "Analysis callback finished [Experiment ID: %s][Analysis Callback ID: %s]",
                self.experiment_id,
                callback_id,
            )
            return callback_id, True
        except Exception as ex:  # pylint: disable=broad-except
            self._analysis_callbacks[callback_id].status = AnalysisStatus.ERROR
            tb_text = "".join(traceback.format_exception(type(ex), ex, ex.__traceback__))
            error_msg = (
                f"Analysis callback failed [Experiment ID: {self.experiment_id}]"
                f"[Analysis Callback ID: {callback_id}]:\n{tb_text}"
            )
            self._analysis_callbacks[callback_id].error_msg = error_msg
            LOG.warning(error_msg)
            return callback_id, False

    def _add_result_data(self, result: Result, job_id: Optional[str] = None) -> None:
        """Add data from a Result object

        Args:
            result: Result object containing data to be added.
            job_id: The id of the job the result came from. If `None`, the
            job id in `result` is used.
        """
        if job_id is None:
            job_id = result.job_id
        if job_id not in self._jobs:
            self._jobs[job_id] = None
            self.job_ids.append(job_id)
        with self._result_data.lock:
            # Lock data while adding all result data
            for i, _ in enumerate(result.results):
                data = result.data(i)
                data["job_id"] = job_id
                if "counts" in data:
                    # Format to Counts object rather than hex dict
                    data["counts"] = result.get_counts(i)
                expr_result = result.results[i]
                if hasattr(expr_result, "header") and hasattr(expr_result.header, "metadata"):
                    data["metadata"] = expr_result.header.metadata
                data["shots"] = expr_result.shots
                data["meas_level"] = expr_result.meas_level
                if hasattr(expr_result, "meas_return"):
                    data["meas_return"] = expr_result.meas_return
                self._result_data.append(data)

    def _retrieve_data(self):
        """Retrieve job data if missing experiment data."""
        # Get job results if missing in experiment data.
        if self.provider is None:
            return
        retrieved_jobs = {}
        jobs_to_retrieve = []  # the list of all jobs to retrieve from the server

        # first find which jobs are listed in the `job_ids` field of the experiment data
        if self.job_ids is not None:
            for jid in self.job_ids:
                if jid not in self._jobs or self._jobs[jid] is None:
                    jobs_to_retrieve.append(jid)

        for jid in jobs_to_retrieve:
            try:
                LOG.debug("Retrieving job [Job ID: %s]", jid)
                job = self.provider.retrieve_job(jid)
                retrieved_jobs[jid] = job
            except Exception:  # pylint: disable=broad-except
                LOG.warning(
                    "Unable to retrieve data from job [Job ID: %s]",
                    jid,
                )
        # Add retrieved job objects to stored jobs and extract data
        for jid, job in retrieved_jobs.items():
            self._jobs[jid] = job
            if job.status() in JOB_FINAL_STATES:
                # Add job results synchronously
                self._add_job_data(job)
            else:
                # Add job results asynchronously
                self._add_job_future(job)

    def data(
        self,
        index: Optional[Union[int, slice, str]] = None,
    ) -> Union[Dict, List[Dict]]:
        """Return the experiment data at the specified index.

        Args:
            index: Index of the data to be returned.
                Several types are accepted for convenience:

                    * None: Return all experiment data.
                    * int: Specific index of the data.
                    * slice: A list slice of data indexes.
                    * str: ID of the job that produced the data.

        Returns:
            Experiment data.

        Raises:
            TypeError: If the input `index` has an invalid type.
        """
        self._retrieve_data()
        if index is None:
            return self._result_data.copy()
        if isinstance(index, (int, slice)):
            return self._result_data[index]
        if isinstance(index, str):
            return [data for data in self._result_data if data.get("job_id") == index]
        raise TypeError(f"Invalid index type {type(index)}.")

    @do_auto_save
    def add_figures(
        self,
        figures: Union[_FigureT, List[_FigureT]],
        figure_names: Optional[Union[str, List[str]]] = None,
        overwrite: bool = False,
        save_figure: Optional[bool] = None,
    ) -> Union[str, List[str]]:
        """Add the experiment figure.

        Args:
            figures: Paths of the figure files or figure data.
            figure_names: Names of the figures. If ``None``, use the figure file
                names, if given, or a generated name of the format ``experiment_type``, figure
                index, first 5 elements of ``device_components``, and first 8 digits of the
                experiment ID connected by underscores, such as ``T1_Q0_0123abcd.svg``. If `figures`
                is a list, then `figure_names` must also be a list of the same length or ``None``.
            overwrite: Whether to overwrite the figure if one already exists with
                the same name. By default, overwrite is ``False`` and the figure will be renamed
                with an incrementing numerical suffix. For example, trying to save ``figure.svg`` when
                ``figure.svg`` already exists will save it as ``figure-1.svg``, and trying to save
                ``figure-1.svg`` when ``figure-1.svg`` already exists will save it as ``figure-2.svg``.
            save_figure: Whether to save the figure in the database. If ``None``,
                the ``auto-save`` attribute is used.

        Returns:
            Figure names in SVG format.

        Raises:
            ValueError: If an input parameter has an invalid value.
        """
        if figure_names is not None and not isinstance(figure_names, list):
            figure_names = [figure_names]
        if not isinstance(figures, list):
            figures = [figures]
        if figure_names is not None and len(figures) != len(figure_names):
            raise ValueError(
                "The parameter figure_names must be None or a list of "
                "the same size as the parameter figures."
            )

        added_figs = []
        for idx, figure in enumerate(figures):
            if figure_names is None:
                if isinstance(figure, str):
                    # figure is a filename, so we use it as the name
                    fig_name = figure
                elif not isinstance(figure, FigureData):
                    # Generate a name in the form StandardRB_Q0_Q1_Q2_b4f1d8ad-1.svg
                    fig_name = (
                        f"{self.experiment_type}_"
                        f'{"_".join(str(i) for i in self.metadata.get("device_components", [])[:5])}_'
                        f"{self.experiment_id[:8]}.svg"
                    )
                else:
                    # Keep the existing figure name if there is one
                    fig_name = figure.name
            else:
                fig_name = figure_names[idx]
            if not fig_name.endswith(".svg"):
                LOG.info("File name %s does not have an SVG extension. A '.svg' is added.")
                fig_name += ".svg"

            existing_figure = fig_name in self._figures
            if existing_figure and not overwrite:
                # Remove any existing suffixes then generate new figure name
                # StandardRB_Q0_Q1_Q2_b4f1d8ad.svg becomes StandardRB_Q0_Q1_Q2_b4f1d8ad
                fig_name_chunked = fig_name.rsplit("-", 1)
                if len(fig_name_chunked) != 1:  # Figure name already has a suffix
                    # This extracts StandardRB_Q0_Q1_Q2_b4f1d8ad as the prefix from
                    # StandardRB_Q0_Q1_Q2_b4f1d8ad-1.svg
                    fig_name_prefix = fig_name_chunked[0]
                    try:
                        fig_name_suffix = int(fig_name_chunked[1].rsplit(".", 1)[0])
                    except ValueError:  # the suffix is not an int, add our own suffix
                        # my-custom-figure-name will be the prefix of my-custom-figure-name.svg
                        fig_name_prefix = fig_name.rsplit(".", 1)[0]
                        fig_name_suffix = 0
                else:
                    # StandardRB_Q0_Q1_Q2_b4f1d8ad.svg has no hyphens so
                    # StandardRB_Q0_Q1_Q2_b4f1d8ad would be its prefix
                    fig_name_prefix = fig_name.rsplit(".", 1)[0]
                    fig_name_suffix = 0
                fig_name = f"{fig_name_prefix}-{fig_name_suffix + 1}.svg"
                while fig_name in self._figures:  # Increment suffix until the name isn't taken
                    # If StandardRB_Q0_Q1_Q2_b4f1d8ad-1.svg already exists,
                    # StandardRB_Q0_Q1_Q2_b4f1d8ad-2.svg will be the name of this figure
                    fig_name_suffix += 1
                    fig_name = f"{fig_name_prefix}-{fig_name_suffix + 1}.svg"

            # figure_data = None
            if isinstance(figure, str):
                with open(figure, "rb") as file:
                    figure = file.read()

            # check whether the figure is already wrapped, meaning it came from a sub-experiment
            if isinstance(figure, FigureData):
                figure_data = figure.copy(new_name=fig_name)
                figure = figure_data.figure

            else:
                figure_metadata = {
                    "qubits": self.metadata.get("physical_qubits"),
                    "device_components": self.metadata.get("device_components"),
                    "experiment_type": self.experiment_type,
                }
                figure_data = FigureData(figure=figure, name=fig_name, metadata=figure_metadata)

            self._figures[fig_name] = figure_data
            self._db_data.figure_names.append(fig_name)

            save = save_figure if save_figure is not None else self.auto_save
            if save and self._service:
                if isinstance(figure, pyplot.Figure):
                    figure = plot_to_svg_bytes(figure)
                self._service.create_or_update_figure(
                    experiment_id=self.experiment_id,
                    figure=figure,
                    figure_name=fig_name,
                    create=not existing_figure,
                )
            added_figs.append(fig_name)

        return added_figs if len(added_figs) != 1 else added_figs[0]

    @do_auto_save
    def delete_figure(
        self,
        figure_key: Union[str, int],
    ) -> str:
        """Add the experiment figure.

        Args:
            figure_key: Name or index of the figure.

        Returns:
            Figure name.

        Raises:
            ExperimentEntryNotFound: If the figure is not found.
        """
        if isinstance(figure_key, int):
            figure_key = self._figures.keys()[figure_key]
        elif figure_key not in self._figures:
            raise ExperimentEntryNotFound(f"Figure {figure_key} not found.")

        del self._figures[figure_key]
        self._deleted_figures.append(figure_key)

        if self._service and self.auto_save:
            with service_exception_to_warning():
                self.service.delete_figure(experiment_id=self.experiment_id, figure_name=figure_key)
            self._deleted_figures.remove(figure_key)

        return figure_key

    def figure(
        self,
        figure_key: Union[str, int],
        file_name: Optional[str] = None,
    ) -> Union[int, FigureData]:
        """Retrieve the specified experiment figure.

        Args:
            figure_key: Name or index of the figure.
            file_name: Name of the local file to save the figure to. If ``None``,
                the content of the figure is returned instead.

        Returns:
            The size of the figure if `file_name` is specified. Otherwise the
            content of the figure as a `FigureData` object.

        Raises:
            ExperimentEntryNotFound: If the figure cannot be found.
        """
        if isinstance(figure_key, int):
            if figure_key < 0 or figure_key >= len(self._figures.keys()):
                raise ExperimentEntryNotFound(f"Figure {figure_key} not found.")
            figure_key = self._figures.keys()[figure_key]

        figure_data = self._figures.get(figure_key, None)
        if figure_data is None and self.service:
            figure = self.service.figure(experiment_id=self.experiment_id, figure_name=figure_key)
            figure_data = FigureData(figure=figure, name=figure_key)
            self._figures[figure_key] = figure_data

        if figure_data is None:
            raise ExperimentEntryNotFound(f"Figure {figure_key} not found.")

        if file_name:
            with open(file_name, "wb") as output:
                num_bytes = output.write(figure_data.figure)
                return num_bytes
        return figure_data

    @deprecate_arg(
        name="results",
        since="0.6",
        additional_msg="Use keyword arguments rather than creating an AnalysisResult object.",
        package_name="qiskit-experiments",
        pending=True,
    )
    @do_auto_save
    def add_analysis_results(
        self,
        results: Optional[Union[AnalysisResult, List[AnalysisResult]]] = None,
        *,
        name: Optional[str] = None,
        value: Optional[Any] = None,
        quality: Optional[str] = None,
        components: Optional[List[DeviceComponent]] = None,
        experiment: Optional[str] = None,
        experiment_id: Optional[str] = None,
        result_id: Optional[str] = None,
        tags: Optional[List[str]] = None,
        backend: Optional[str] = None,
        run_time: Optional[datetime] = None,
        created_time: Optional[datetime] = None,
        **extra_values,
    ) -> None:
        """Save the analysis result.

        Args:
            results: Analysis results to be saved.
            name: Name of the result entry.
            value: Analyzed quantity.
            quality: Quality of the data.
            components: Associated device components.
            experiment: String identifier of the associated experiment.
            experiment_id: ID of the associated experiment.
            result_id: ID of this analysis entry. If not set a random UUID is generated.
            tags: List of arbitrary tags.
            backend: Name of associated backend.
            run_time: The date time when the experiment started to run on the device.
            created_time: The date time when this analysis is performed.
            extra_values: Arbitrary keyword arguments for supplementary information.
                New dataframe columns are created in the analysis result table with added keys.
        """
        if results is not None:
            # TODO deprecate this path
            if not isinstance(results, list):
                results = [results]
            for result in results:
                extra_values = result.extra.copy()
                if result.chisq is not None:
                    # Move chisq to extra.
                    # This is not global outcome, e.g. QPT doesn't provide chisq.
                    extra_values["chisq"] = result.chisq
                experiment = extra_values.pop("experiment", self.experiment_type)
                backend = extra_values.pop("backend", self.backend_name)
                run_time = extra_values.pop("run_time", self.running_time)
                created_time = extra_values.pop("created_time", None)
                self._analysis_results.add_entry(
                    name=result.name,
                    value=result.value,
                    quality=result.quality,
                    components=result.device_components,
                    experiment=experiment,
                    experiment_id=result.experiment_id,
                    result_id=result.result_id,
                    tags=result.tags,
                    backend=backend,
                    run_time=run_time,
                    created_time=created_time,
                    **extra_values,
                )
                if self.auto_save:
                    result.save()
        else:
            experiment = experiment or self.experiment_type
            experiment_id = experiment_id or self.experiment_id
            tags = tags or []
            backend = backend or self.backend_name

            self._analysis_results.add_entry(
                result_id=result_id,
                name=name,
                value=value,
                quality=quality,
                components=components,
                experiment=experiment,
                experiment_id=experiment_id,
                tags=tags or [],
                backend=backend,
                run_time=run_time,  # TODO add job RUNNING time
                created_time=created_time,
                **extra_values,
            )
            if self.auto_save:
                last_index = self._analysis_results.result_ids()[-1][:8]
                service_result = _series_to_service_result(
                    series=self._analysis_results.get_entry(last_index),
                    service=self._service,
                    auto_save=False,
                )
                service_result.save()

    @do_auto_save
    def delete_analysis_result(
        self,
        result_key: Union[int, str],
    ) -> str:
        """Delete the analysis result.

        Args:
            result_key: ID or index of the analysis result to be deleted.

        Returns:
            Analysis result ID.

        Raises:
            ExperimentEntryNotFound: If analysis result not found or multiple entries are found.
        """
        # Retrieve from DB if needed.
        to_delete = self.analysis_results(
            index=result_key,
            block=False,
            columns="all",
            dataframe=True,
        )
        if not isinstance(to_delete, pd.Series):
            raise ExperimentEntryNotFound(
                f"Multiple entries are found with result_key = {result_key}. "
                "Try another key that can uniquely determine entry to delete."
            )

        self._analysis_results.drop_entry(str(to_delete.name))
        if self._service and self.auto_save:
            with service_exception_to_warning():
                self.service.delete_analysis_result(result_id=to_delete.result_id)
        else:
            self._deleted_analysis_results.append(to_delete.result_id)

        return to_delete.result_id

    def _retrieve_analysis_results(self, refresh: bool = False):
        """Retrieve service analysis results.

        Args:
            refresh: Retrieve the latest analysis results from the server, if
                an experiment service is available.
        """
        # Get job results if missing experiment data.
        if self.service and (len(self._analysis_results) == 0 or refresh):
            retrieved_results = self.service.analysis_results(
                experiment_id=self.experiment_id, limit=None, json_decoder=self._json_decoder
            )
            for result in retrieved_results:
                # Canonicalize IBM specific data structure.
                # TODO define proper data schema on frontend and delegate this to service.
                cano_quality = AnalysisResult.RESULT_QUALITY_TO_TEXT.get(result.quality, "unknown")
                cano_components = [to_component(c) for c in result.device_components]
                extra = result.result_data["_extra"]
                if result.chisq is not None:
                    extra["chisq"] = result.chisq
                self._analysis_results.add_entry(
                    name=result.result_type,
                    value=result.result_data["_value"],
                    quality=cano_quality,
                    components=cano_components,
                    experiment_id=result.experiment_id,
                    result_id=result.result_id,
                    tags=result.tags,
                    backend=result.backend_name,
                    created_time=result.creation_datetime,
                    **extra,
                )

    @deprecate_arg(
        name="dataframe",
        deprecation_description="Setting ``dataframe`` to False in analysis_results",
        since="0.6",
        package_name="qiskit-experiments",
        pending=True,
        predicate=lambda dataframe: not dataframe,
    )
    def analysis_results(
        self,
        index: Optional[Union[int, slice, str]] = None,
        refresh: bool = False,
        block: bool = True,
        timeout: Optional[float] = None,
        columns: Union[str, List[str]] = "default",
        dataframe: bool = False,
    ) -> Union[AnalysisResult, List[AnalysisResult], pd.DataFrame, pd.Series]:
        """Return analysis results associated with this experiment.

        Args:
            index: Index of the analysis result to be returned.
                Several types are accepted for convenience:

                    * None: Return all analysis results.
                    * int: Specific index of the analysis results.
                    * slice: A list slice of indexes.
                    * str: ID or name of the analysis result.

            refresh: Retrieve the latest analysis results from the server, if
                an experiment service is available.
            block: If True block for any analysis callbacks to finish running.
            timeout: max time in seconds to wait for analysis callbacks to finish running.
            columns: Specifying a set of columns to return. You can pass a list of each
                column name to return, otherwise builtin column groups are available.

                    * "all": Return all columns, including metadata to communicate
                        with experiment service, such as entry IDs.
                    * "default": Return columns including analysis result with supplementary
                        information about experiment.
                    * "minimal": Return only analysis subroutine returns.

            dataframe: Set True to return analysis results in the dataframe format.

        Returns:
            Analysis results for this experiment.

        Raises:
            ExperimentEntryNotFound: If the entry cannot be found.
        """
        if block:
            self._wait_for_futures(
                self._analysis_futures.values(), name="analysis", timeout=timeout
            )
        self._retrieve_analysis_results(refresh=refresh)

        out = self._analysis_results.copy()

        if index is not None:
            out = _filter_analysis_results(index, out)
            if out is None:
                msg = [f"Analysis result {index} not found."]
                errors = self.errors()
                if errors:
                    msg.append(f"Errors: {errors}")
                raise ExperimentEntryNotFound("\n".join(msg))

        if dataframe:
            valid_columns = self._analysis_results.filter_columns(columns)
            out = out[valid_columns]
            if len(out) == 1 and index is not None:
                # For backward compatibility.
                # One can directly access attributes with Series. e.g. out.value
                return out.iloc[0]
            return out

        # Convert back into List[AnalysisResult] which is payload for IBM experiment service.
        # This will be removed in future version.
        service_results = []
        for _, series in out.iterrows():
            service_results.append(
                _series_to_service_result(
                    series=series,
                    service=self._service,
                    auto_save=self._auto_save,
                )
            )
        if len(service_results) == 1 and index is not None:
            return service_results[0]
        return service_results

    # Save and load from the database

    def save_metadata(self) -> None:
        """Save this experiments metadata to a database service.

        .. note::
            This method does not save analysis results nor figures.
            Use :meth:`save` for general saving of all experiment data.

            See :meth:`qiskit.providers.experiment.IBMExperimentService.create_experiment`
            for fields that are saved.
        """
        self._save_experiment_metadata()
        for data in self.child_data():
            data.save_metadata()

    def _save_experiment_metadata(self, suppress_errors: bool = True) -> None:
        """Save this experiments metadata to a database service.
        Args:
            suppress_errors: should the method catch exceptions (true) or
            pass them on, potentially aborting the experiment (false)
        Raises:
            ExperimentDataSaveFailed: If the save to the database failed.

        .. note::
            This method does not save analysis results or figures.
            Use :meth:`save` for general saving of all experiment data.

            See :meth:`qiskit.providers.experiment.IBMExperimentService.create_experiment`
            for fields that are saved.
        """
        if not self._service:
            LOG.warning(
                "Experiment cannot be saved because no experiment service is available. "
                "An experiment service is available, for example, "
                "when using an IBM Quantum backend."
            )
            if suppress_errors:
                return
            else:
                raise ExperimentDataSaveFailed("No service found")
        try:
            handle_metadata_separately = self._metadata_too_large()
            if handle_metadata_separately:
                metadata = self._db_data.metadata
                self._db_data.metadata = {}

            result = self.service.create_or_update_experiment(
                self._db_data, json_encoder=self._json_encoder, create=not self._created_in_db
            )
            if isinstance(result, dict):
                created_datetime = result.get("created_at", None)
                updated_datetime = result.get("updated_at", None)
                self._db_data.creation_datetime = parse_utc_datetime(created_datetime)
                self._db_data.updated_datetime = parse_utc_datetime(updated_datetime)

            self._created_in_db = True

            if handle_metadata_separately:
                self.service.file_upload(
                    self._db_data.experiment_id,
                    self._metadata_filename,
                    metadata,
                    json_encoder=self._json_encoder,
                )
                self._db_data.metadata = metadata

        except Exception as ex:  # pylint: disable=broad-except
            # Don't automatically fail the experiment just because its data cannot be saved.
            LOG.error("Unable to save the experiment data: %s", traceback.format_exc())
            if not suppress_errors:
                raise ExperimentDataSaveFailed(
                    f"Experiment data save failed\nError Message:\n{str(ex)}"
                ) from ex

    def _metadata_too_large(self):
        """Determines whether the metadata should be stored in a separate file"""
        # currently the entire POST JSON request body is limited by default to 100kb
        total_metadata_size = sys.getsizeof(json.dumps(self.metadata, cls=self._json_encoder))
        return total_metadata_size > 10000

    def save(
        self,
        suppress_errors: bool = True,
        max_workers: int = 3,
        save_figures: bool = True,
        save_children: bool = True,
    ) -> None:
        """Save the experiment data to a database service.

        Args:
            suppress_errors: should the method catch exceptions (true) or
            pass them on, potentially aborting the experiment (false)
            max_workers: Maximum number of concurrent worker threads (capped by 10)
            save_figures: Whether to save figures in the database or not
            save_children: For composite experiments, whether to save children as well

        Raises:
            ExperimentDataSaveFailed: If no experiment database service
            was found, or the experiment service failed to save

        .. note::
            This saves the experiment metadata, all analysis results, and all
            figures. Depending on the number of figures and analysis results this
            operation could take a while.

            To only update a previously saved experiments metadata (eg for
            additional tags or notes) use :meth:`save_metadata`.
        """
        # TODO - track changes
        if not self._service:
            LOG.warning(
                "Experiment cannot be saved because no experiment service is available. "
                "An experiment service is available, for example, "
                "when using an IBM Quantum backend."
            )
            if suppress_errors:
                return
            else:
                raise ExperimentDataSaveFailed("No service found")
        if max_workers > self._max_workers_cap:
            LOG.warning(
                "max_workers cannot be larger than %s. Setting max_workers = %s now.",
                self._max_workers_cap,
                self._max_workers_cap,
            )
            max_workers = self._max_workers_cap
        self._save_experiment_metadata(suppress_errors=suppress_errors)
        if not self._created_in_db:
            LOG.warning("Could not save experiment metadata to DB, aborting experiment save")
            return

        analysis_results_to_create = []
        for _, series in self._analysis_results.copy().iterrows():
            # TODO We should support saving entire dataframe
            #  Calling API per entry takes huge amount of time.
            legacy_result = _series_to_service_result(
                series=series,
                service=self._service,
                auto_save=False,
            )
            analysis_results_to_create.append(legacy_result._db_data)
        try:
            self.service.create_analysis_results(
                data=analysis_results_to_create,
                blocking=True,
                json_encoder=self._json_encoder,
                max_workers=max_workers,
            )
        except Exception as ex:  # pylint: disable=broad-except
            # Don't automatically fail the experiment just because its data cannot be saved.
            LOG.error("Unable to save the experiment data: %s", traceback.format_exc())
            if not suppress_errors:
                raise ExperimentDataSaveFailed(
                    f"Analysis result save failed\nError Message:\n{str(ex)}"
                ) from ex

        for result in self._deleted_analysis_results.copy():
            with service_exception_to_warning():
                self._service.delete_analysis_result(result_id=result)
            self._deleted_analysis_results.remove(result)

        if save_figures:
            with self._figures.lock:
                figures_to_create = []
                for name, figure in self._figures.items():
                    if figure is None:
                        continue
                    # currently only the figure and its name are stored in the database
                    if isinstance(figure, FigureData):
                        figure = figure.figure
                        LOG.debug("Figure metadata is currently not saved to the database")
                    if isinstance(figure, pyplot.Figure):
                        figure = plot_to_svg_bytes(figure)
                    figures_to_create.append((figure, name))
                try:
                    self.service.create_figures(
                        experiment_id=self.experiment_id,
                        figure_list=figures_to_create,
                        blocking=True,
                        max_workers=max_workers,
                    )
                except Exception as ex:  # pylint: disable=broad-except
                    LOG.error("Unable to save figures: %s", traceback.format_exc())
                    if not suppress_errors:
                        raise ExperimentDataSaveFailed(
                            f"Figure save failed\nError Message:\n{str(ex)}"
                        ) from ex

        for name in self._deleted_figures.copy():
            with service_exception_to_warning():
                self._service.delete_figure(experiment_id=self.experiment_id, figure_name=name)
            self._deleted_figures.remove(name)

        if not self.service.local and self.verbose:
            print(
                "You can view the experiment online at "
                f"https://quantum-computing.ibm.com/experiments/{self.experiment_id}"
            )
        # handle children, but without additional prints
        if save_children:
            for data in self._child_data.values():
                original_verbose = data.verbose
                data.verbose = False
                data.save(
                    suppress_errors=suppress_errors,
                    max_workers=max_workers,
                    save_figures=save_figures,
                )
                data.verbose = original_verbose

    def jobs(self) -> List[Job]:
        """Return a list of jobs for the experiment"""
        return self._jobs.values()

    def cancel_jobs(self, ids: Optional[Union[str, List[str]]] = None) -> bool:
        """Cancel any running jobs.

        Args:
            ids: Job(s) to cancel. If None all non-finished jobs will be cancelled.

        Returns:
            True if the specified jobs were successfully cancelled
            otherwise false.
        """
        if isinstance(ids, str):
            ids = [ids]

        with self._jobs.lock:
            all_cancelled = True
            for jid, job in reversed(self._jobs.items()):
                if ids and jid not in ids:
                    # Skip cancelling this callback
                    continue
                if job and job.status() not in JOB_FINAL_STATES:
                    try:
                        job.cancel()
                        LOG.warning("Cancelled job [Job ID: %s]", jid)
                    except Exception as err:  # pylint: disable=broad-except
                        all_cancelled = False
                        LOG.warning("Unable to cancel job [Job ID: %s]:\n%s", jid, err)
                        continue

                # Remove done or cancelled job futures
                if jid in self._job_futures:
                    del self._job_futures[jid]

        return all_cancelled

    def cancel_analysis(self, ids: Optional[Union[str, List[str]]] = None) -> bool:
        """Cancel any queued analysis callbacks.

        .. note::
            A currently running analysis callback cannot be cancelled.

        Args:
            ids: Analysis callback(s) to cancel. If None all non-finished
                 analysis will be cancelled.

        Returns:
            True if the specified analysis callbacks were successfully
            cancelled otherwise false.
        """
        if isinstance(ids, str):
            ids = [ids]

        # Lock analysis futures so we can't add more while trying to cancel
        with self._analysis_futures.lock:
            all_cancelled = True
            not_running = []
            for cid, callback in reversed(self._analysis_callbacks.items()):
                if ids and cid not in ids:
                    # Skip cancelling this callback
                    continue

                # Set event to cancel callback
                callback.event.set()

                # Check for running callback that can't be cancelled
                if callback.status == AnalysisStatus.RUNNING:
                    all_cancelled = False
                    LOG.warning(
                        "Unable to cancel running analysis callback [Experiment ID: %s]"
                        "[Analysis Callback ID: %s]",
                        self.experiment_id,
                        cid,
                    )
                else:
                    not_running.append(cid)

            # Wait for completion of other futures cancelled via event.set
            waited = futures.wait([self._analysis_futures[cid] for cid in not_running], timeout=1)
            # Get futures that didn't raise exception
            for fut in waited.done:
                if fut.done() and not fut.exception():
                    cid = fut.result()[0]
                    if cid in self._analysis_futures:
                        del self._analysis_futures[cid]

        return all_cancelled

    def cancel(self) -> bool:
        """Attempt to cancel any running jobs and queued analysis callbacks.

        .. note::
            A running analysis callback cannot be cancelled.

        Returns:
            True if all jobs and analysis are successfully cancelled, otherwise false.
        """
        # Cancel analysis first since it is queued on jobs, then cancel jobs
        # otherwise there can be a race issue when analysis starts running
        # as soon as jobs are cancelled
        analysis_cancelled = self.cancel_analysis()
        jobs_cancelled = self.cancel_jobs()
        return analysis_cancelled and jobs_cancelled

    def block_for_results(self, timeout: Optional[float] = None) -> "ExperimentData":
        """Block until all pending jobs and analysis callbacks finish.

        Args:
            timeout: Timeout in seconds for waiting for results.

        Returns:
            The experiment data with finished jobs and post-processing.
        """
        start_time = time.time()
        with self._job_futures.lock and self._analysis_futures.lock:
            # Lock threads to get all current job and analysis futures
            # at the time of function call and then release the lock
            job_ids = self._job_futures.keys()
            job_futs = self._job_futures.values()
            analysis_ids = self._analysis_futures.keys()
            analysis_futs = self._analysis_futures.values()

        # Wait for futures
        self._wait_for_futures(job_futs + analysis_futs, name="jobs and analysis", timeout=timeout)
        # Clean up done job futures
        num_jobs = len(job_ids)
        for jid, fut in zip(job_ids, job_futs):
            if (fut.done() and not fut.exception()) or fut.cancelled():
                if jid in self._job_futures:
                    del self._job_futures[jid]
                    num_jobs -= 1

        # Clean up done analysis futures
        num_analysis = len(analysis_ids)
        for cid, fut in zip(analysis_ids, analysis_futs):
            if (fut.done() and not fut.exception()) or fut.cancelled():
                if cid in self._analysis_futures:
                    del self._analysis_futures[cid]
                    num_analysis -= 1

        # Check if more futures got added while this function was running
        # and block recursively. This could happen if an analysis callback
        # spawns another callback or creates more jobs
        if len(self._job_futures) > num_jobs or len(self._analysis_futures) > num_analysis:
            time_taken = time.time() - start_time
            if timeout is not None:
                timeout = max(0, timeout - time_taken)
            return self.block_for_results(timeout=timeout)

        return self

    def _wait_for_futures(
        self, futs: List[futures.Future], name: str = "futures", timeout: Optional[float] = None
    ) -> bool:
        """Wait for jobs to finish running.

        Args:
            futs: Job or analysis futures to wait for.
            name: type name for future for logger messages.
            timeout: The length of time to wait for all jobs
                     before returning False.

        Returns:
            True if all jobs finished. False if timeout time was reached
            or any jobs were cancelled or had an exception.
        """
        waited = futures.wait(futs, timeout=timeout)
        value = True

        # Log futures still running after timeout
        if waited.not_done:
            LOG.info(
                "Waiting for %s timed out before completion [Experiment ID: %s].",
                name,
                self.experiment_id,
            )
            value = False

        # Check for futures that were cancelled or errored
        excepts = ""
        for fut in waited.done:
            ex = fut.exception()
            if ex:
                excepts += "\n".join(traceback.format_exception(type(ex), ex, ex.__traceback__))
                value = False
            elif fut.cancelled():
                LOG.debug(
                    "%s was cancelled before completion [Experiment ID: %s]",
                    name,
                    self.experiment_id,
                )
                value = False
            elif not fut.result()[1]:
                # The job/analysis did not succeed, and the failure reflects in the second
                # returned value of _add_job_data/_run_analysis_callback. See details in Issue #866.
                value = False
        if excepts:
            LOG.error(
                "%s raised exceptions [Experiment ID: %s]:%s", name, self.experiment_id, excepts
            )

        return value

    def status(self) -> ExperimentStatus:
        """Return the experiment status.

        Possible return values for :class:`.ExperimentStatus` are

        * :attr:`~.ExperimentStatus.EMPTY` - experiment data is empty
        * :attr:`~.ExperimentStatus.INITIALIZING` - experiment jobs are being initialized
        * :attr:`~.ExperimentStatus.QUEUED` - experiment jobs are queued
        * :attr:`~.ExperimentStatus.RUNNING` - experiment jobs is actively running
        * :attr:`~.ExperimentStatus.CANCELLED` - experiment jobs or analysis has been cancelled
        * :attr:`~.ExperimentStatus.POST_PROCESSING` - experiment analysis is actively running
        * :attr:`~.ExperimentStatus.DONE` - experiment jobs and analysis have successfully run
        * :attr:`~.ExperimentStatus.ERROR` - experiment jobs or analysis incurred an error

        .. note::

            If an experiment has status :attr:`~.ExperimentStatus.ERROR`
            there may still be pending or running jobs. In these cases it
            may be beneficial to call :meth:`cancel_jobs` to terminate these
            remaining jobs.

        Returns:
            The experiment status.
        """
        if all(
            len(container) == 0
            for container in [
                self._result_data,
                self._jobs,
                self._job_futures,
                self._analysis_callbacks,
                self._analysis_futures,
                self._figures,
                self._analysis_results,
            ]
        ):
            return ExperimentStatus.EMPTY

        # Return job status is job is not DONE
        try:
            return {
                JobStatus.INITIALIZING: ExperimentStatus.INITIALIZING,
                JobStatus.QUEUED: ExperimentStatus.QUEUED,
                JobStatus.VALIDATING: ExperimentStatus.VALIDATING,
                JobStatus.RUNNING: ExperimentStatus.RUNNING,
                JobStatus.CANCELLED: ExperimentStatus.CANCELLED,
                JobStatus.ERROR: ExperimentStatus.ERROR,
            }[self.job_status()]
        except KeyError:
            pass

        # Return analysis status if Done, cancelled or error
        try:
            return {
                AnalysisStatus.DONE: ExperimentStatus.DONE,
                AnalysisStatus.CANCELLED: ExperimentStatus.CANCELLED,
                AnalysisStatus.ERROR: ExperimentStatus.ERROR,
            }[self.analysis_status()]
        except KeyError:
            return ExperimentStatus.POST_PROCESSING

    def job_status(self) -> JobStatus:
        """Return the experiment job execution status.

        Possible return values for :class:`qiskit.providers.jobstatus.JobStatus` are

        * ``ERROR`` - if any job incurred an error
        * ``CANCELLED`` - if any job is cancelled.
        * ``RUNNING`` - if any job is still running.
        * ``QUEUED`` - if any job is queued.
        * ``VALIDATING`` - if any job is being validated.
        * ``INITIALIZING`` - if any job is being initialized.
        * ``DONE`` - if all jobs are finished.

        .. note::

            If an experiment has status ``ERROR`` or ``CANCELLED`` there may still be
            pending or running jobs. In these cases it may be beneficial to call
            :meth:`cancel_jobs` to terminate these remaining jobs.

        Returns:
            The job execution status.
        """
        statuses = set()
        with self._jobs.lock:

            # No jobs present
            if not self._jobs:
                return JobStatus.DONE

            statuses = set()
            for job in self._jobs.values():
                if job:
                    statuses.add(job.status())

        # If any jobs are in non-DONE state return that state
        for stat in [
            JobStatus.ERROR,
            JobStatus.CANCELLED,
            JobStatus.RUNNING,
            JobStatus.QUEUED,
            JobStatus.VALIDATING,
            JobStatus.INITIALIZING,
        ]:
            if stat in statuses:
                return stat

        return JobStatus.DONE

    def analysis_status(self) -> AnalysisStatus:
        """Return the data analysis post-processing status.

        Possible return values for :class:`.AnalysisStatus` are

        * :attr:`~.AnalysisStatus.ERROR` - if any analysis callback incurred an error
        * :attr:`~.AnalysisStatus.CANCELLED` - if any analysis callback is cancelled.
        * :attr:`~.AnalysisStatus.RUNNING` - if any analysis callback is actively running.
        * :attr:`~.AnalysisStatus.QUEUED` - if any analysis callback is queued.
        * :attr:`~.AnalysisStatus.DONE` - if all analysis callbacks have successfully run.

        Returns:
            Then analysis status.
        """
        statuses = set()
        for status in self._analysis_callbacks.values():
            statuses.add(status.status)

        for stat in [
            AnalysisStatus.ERROR,
            AnalysisStatus.CANCELLED,
            AnalysisStatus.RUNNING,
            AnalysisStatus.QUEUED,
        ]:
            if stat in statuses:
                return stat

        return AnalysisStatus.DONE

    def job_errors(self) -> str:
        """Return any errors encountered in job execution."""
        errors = []

        # Get any job errors
        for job in self._jobs.values():
            if job and job.status() == JobStatus.ERROR:
                if hasattr(job, "error_message"):
                    error_msg = job.error_message()
                else:
                    error_msg = ""
                errors.append(f"\n[Job ID: {job.job_id()}]: {error_msg}")

        # Get any job futures errors:
        for jid, fut in self._job_futures.items():
            if fut and fut.done() and fut.exception():
                ex = fut.exception()
                errors.append(
                    f"[Job ID: {jid}]"
                    "\n".join(traceback.format_exception(type(ex), ex, ex.__traceback__))
                )
        return "".join(errors)

    def analysis_errors(self) -> str:
        """Return any errors encountered during analysis callbacks."""
        errors = []

        # Get any callback errors
        for cid, callback in self._analysis_callbacks.items():
            if callback.status == AnalysisStatus.ERROR:
                errors.append(f"\n[Analysis Callback ID: {cid}]: {callback.error_msg}")

        return "".join(errors)

    def errors(self) -> str:
        """Return errors encountered during job and analysis execution.

        .. note::
            To display only job or analysis errors use the
            :meth:`job_errors` or :meth:`analysis_errors` methods.

        Returns:
            Experiment errors.
        """
        return self.job_errors() + self.analysis_errors()

    # Children handling

    def add_child_data(self, experiment_data: ExperimentData):
        """Add child experiment data to the current experiment data"""
        experiment_data.parent_id = self.experiment_id
        self._child_data[experiment_data.experiment_id] = experiment_data
        self.metadata["child_data_ids"] = self._child_data.keys()

    def child_data(
        self, index: Optional[Union[int, slice, str]] = None
    ) -> Union[ExperimentData, List[ExperimentData]]:
        """Return child experiment data.

        Args:
            index: Index of the child experiment data to be returned.
                Several types are accepted for convenience:

                    * None: Return all child data.
                    * int: Specific index of the child data.
                    * slice: A list slice of indexes.
                    * str: experiment ID of the child data.

        Returns:
            The requested single or list of child experiment data.

        Raises:
            QiskitError: If the index or ID of the child experiment data
                         cannot be found.
        """
        if index is None:
            return self._child_data.values()
        if isinstance(index, (int, slice)):
            return self._child_data.values()[index]
        if isinstance(index, str):
            return self._child_data[index]
        raise QiskitError(f"Invalid index type {type(index)}.")

    @classmethod
    def load(
        cls,
        experiment_id: str,
        service: Optional[IBMExperimentService] = None,
        provider: Optional[Provider] = None,
    ) -> "ExperimentData":
        """Load a saved experiment data from a database service.

        Args:
            experiment_id: Experiment ID.
            service: the database service.
            provider: an IBMProvider required for loading job data and
            can be used to initialize the service.

        Returns:
            The loaded experiment data.
        Raises:
            ExperimentDataError: If not service nor provider were given.
        """
        if service is None:
            if provider is None:
                raise ExperimentDataError(
                    "Loading an experiment requires a valid ibm provider or experiment service"
                )
            service = cls.get_service_from_provider(provider)
        data = service.experiment(experiment_id, json_decoder=cls._json_decoder)
        if service.experiment_has_file(experiment_id, cls._metadata_filename):
            metadata = service.file_download(
                experiment_id, cls._metadata_filename, json_decoder=cls._json_decoder
            )
            data.metadata.update(metadata)
        expdata = cls(service=service, db_data=data, provider=provider)

        # Retrieve data and analysis results
        # Maybe this isn't necessary but the repr of the class should
        # be updated to show correct number of results including remote ones
        expdata._retrieve_data()
        expdata._retrieve_analysis_results()

        # mark it as existing in the DB
        expdata._created_in_db = True

        child_data_ids = expdata.metadata.pop("child_data_ids", [])
        child_data = [
            ExperimentData.load(child_id, service, provider) for child_id in child_data_ids
        ]
        expdata._set_child_data(child_data)

        return expdata

    def copy(self, copy_results: bool = True) -> "ExperimentData":
        """Make a copy of the experiment data with a new experiment ID.

        Args:
            copy_results: If True copy the analysis results and figures
                          into the returned container, along with the
                          experiment data and metadata. If False only copy
                          the experiment data and metadata.

        Returns:
            A copy of the experiment data object with the same data
            but different IDs.

        .. note:
            If analysis results and figures are copied they will also have
            new result IDs and figure names generated for the copies.

            This method can not be called from an analysis callback. It waits
            for analysis callbacks to complete before copying analysis results.
        """
        new_instance = ExperimentData(
            backend=self.backend,
            service=self.service,
            parent_id=self.parent_id,
            job_ids=self.job_ids,
            child_data=list(self._child_data.values()),
            verbose=self.verbose,
        )
        new_instance._db_data = self._db_data.copy()
        new_instance._db_data.experiment_id = str(
            uuid.uuid4()
        )  # different id for copied experiment
        if self.experiment is None:
            new_instance._experiment = None
        else:
            new_instance._experiment = self.experiment.copy()

        LOG.debug(
            "Copying experiment data [Experiment ID: %s]: %s",
            self.experiment_id,
            new_instance.experiment_id,
        )

        # Copy basic properties and metadata

        new_instance._jobs = self._jobs.copy_object()
        new_instance._auto_save = self._auto_save
        new_instance._extra_data = self._extra_data

        # Copy circuit result data and jobs
        with self._result_data.lock:  # Hold the lock so no new data can be added.
            new_instance._result_data = self._result_data.copy_object()
            for jid, fut in self._job_futures.items():
                if not fut.done():
                    new_instance._add_job_future(new_instance._jobs[jid])

        # If not copying results return the object
        if not copy_results:
            return new_instance

        # Copy results and figures.
        # This requires analysis callbacks to finish
        self._wait_for_futures(self._analysis_futures.values(), name="analysis")
        new_instance._analysis_results = self._analysis_results.copy_object()
        with self._figures.lock:
            new_instance._figures = ThreadSafeOrderedDict()
            new_instance.add_figures(self._figures.values())

        # Recursively copy child data
        child_data = [data.copy(copy_results=copy_results) for data in self.child_data()]
        new_instance._set_child_data(child_data)
        return new_instance

    def _set_child_data(self, child_data: List[ExperimentData]):
        """Set child experiment data for the current experiment."""
        self._child_data = ThreadSafeOrderedDict()
        for data in child_data:
            self.add_child_data(data)
        self._db_data.metadata["child_data_ids"] = self._child_data.keys()

    def _set_service(self, service: IBMExperimentService, replace: bool = None) -> None:
        """Set the service to be used for storing experiment data,
           to this experiment itself and its descendants.

        Args:
            service: Service to be used.
            replace: Should an existing service be replaced?
            If not, and a current service exists, exception is raised

        Raises:
            ExperimentDataError: If an experiment service is already being used and `replace==False`.
        """
        if self._service and not replace:
            raise ExperimentDataError("An experiment service is already being used.")
        self._service = service
        with contextlib.suppress(Exception):
            self.auto_save = self._service.options.get("auto_save", False)
        for data in self.child_data():
            data._set_service(service)

    def add_tags_recursive(self, tags2add: List[str]) -> None:
        """Add tags to this experiment itself and its descendants

        Args:
            tags2add - the tags that will be added to the existing tags
        """
        self.tags += tags2add
        for data in self._child_data.values():
            data.add_tags_recursive(tags2add)

    def remove_tags_recursive(self, tags2remove: List[str]) -> None:
        """Remove tags from this experiment itself and its descendants

        Args:
            tags2remove - the tags that will be removed from the existing tags
        """
        self.tags = [x for x in self.tags if x not in tags2remove]
        for data in self._child_data.values():
            data.remove_tags_recursive(tags2remove)

    # representation and serialization

    def __repr__(self):
        out = f"{type(self).__name__}({self.experiment_type}"
        out += f", {self.experiment_id}"
        if self.parent_id:
            out += f", parent_id={self.parent_id}"
        if self.tags:
            out += f", tags={self.tags}"
        if self.job_ids:
            out += f", job_ids={self.job_ids}"
        if self.share_level:
            out += f", share_level={self.share_level}"
        if self.metadata:
            out += f", metadata=<{len(self.metadata)} items>"
        if self.figure_names:
            out += f", figure_names={self.figure_names}"
        if self.notes:
            out += f", notes={self.notes}"
        if self._extra_data:
            for key, val in self._extra_data.items():
                out += f", {key}={repr(val)}"
        out += ")"
        return out

    def __getattr__(self, name: str) -> Any:
        try:
            return self._extra_data[name]
        except KeyError:
            # pylint: disable=raise-missing-from
            raise AttributeError(f"Attribute {name} is not defined")

    def _safe_serialize_jobs(self):
        """Return serializable object for stored jobs"""
        # Since Job objects are not serializable this removes
        # them from the jobs dict and returns {job_id: None}
        # that can be used to retrieve jobs from a service after loading
        jobs = ThreadSafeOrderedDict()
        with self._jobs.lock:
            for jid in self._jobs.keys():
                jobs[jid] = None
        return jobs

    def _safe_serialize_figures(self):
        """Return serializable object for stored figures"""
        # Convert any MPL figures into SVG images before serializing
        figures = ThreadSafeOrderedDict()
        with self._figures.lock:
            for name, figure in self._figures.items():
                if isinstance(figure, pyplot.Figure):
                    figures[name] = plot_to_svg_bytes(figure)
                else:
                    figures[name] = figure
        return figures

    def __json_encode__(self):
        if any(not fut.done() for fut in self._job_futures.values()):
            raise QiskitError(
                "Not all experiment jobs have finished. Jobs must be "
                "cancelled or done to serialize experiment data."
            )
        if any(not fut.done() for fut in self._analysis_futures.values()):
            raise QiskitError(
                "Not all experiment analysis has finished. Analysis must be "
                "cancelled or done to serialize experiment data."
            )
        json_value = {
            "_db_data": self._db_data,
            "_analysis_results": self._analysis_results,
            "_analysis_callbacks": self._analysis_callbacks,
            "_deleted_figures": self._deleted_figures,
            "_deleted_analysis_results": self._deleted_analysis_results,
            "_result_data": self._result_data,
            "_extra_data": self._extra_data,
            "_created_in_db": self._created_in_db,
            "_figures": self._safe_serialize_figures(),  # Convert figures to SVG
            "_jobs": self._safe_serialize_jobs(),  # Handle non-serializable objects
            "_experiment": self._experiment,
            "_child_data": self._child_data,
            "_running_time": self._running_time,
        }
        # the attribute self._service in charge of the connection and communication with the
        #  experiment db. It doesn't have meaning in the json format so there is no need to serialize
        #  it.
        for att in ["_service", "_backend"]:
            json_value[att] = None
            value = getattr(self, att)
            if value is not None:
                LOG.info("%s cannot be JSON serialized", str(type(value)))

        return json_value

    @classmethod
    def __json_decode__(cls, value):
        ret = cls()
        for att, att_val in value.items():
            setattr(ret, att, att_val)
        return ret

    def __getstate__(self):
        if any(not fut.done() for fut in self._job_futures.values()):
            LOG.warning(
                "Not all job futures have finished."
                " Data from running futures will not be serialized."
            )
        if any(not fut.done() for fut in self._analysis_futures.values()):
            LOG.warning(
                "Not all analysis callbacks have finished."
                " Results from running callbacks will not be serialized."
            )

        state = self.__dict__.copy()

        # Remove non-pickleable attributes
        for key in ["_job_futures", "_analysis_futures", "_analysis_executor", "_monitor_executor"]:
            del state[key]

        # Convert figures to SVG
        state["_figures"] = self._safe_serialize_figures()

        # Handle partially pickleable attributes
        state["_jobs"] = self._safe_serialize_jobs()

        return state

    @staticmethod
    def get_service_from_backend(backend):
        """Initializes the service from the backend data"""
        return ExperimentData.get_service_from_provider(backend.provider)

    @staticmethod
    def get_service_from_provider(provider):
        """Initializes the service from the provider data"""
        db_url = "https://auth.quantum-computing.ibm.com/api"
        try:
            # qiskit-ibmq-provider style
            if hasattr(provider, "credentials"):
                token = provider.credentials.token
            # qiskit-ibm-provider style
            if hasattr(provider, "_account"):
                token = provider._account.token
            service = IBMExperimentService(token=token, url=db_url)
            return service
        except Exception:  # pylint: disable=broad-except
            return None

    def __setstate__(self, state):
        self.__dict__.update(state)
        # Initialize non-pickled attributes
        self._job_futures = ThreadSafeOrderedDict()
        self._analysis_futures = ThreadSafeOrderedDict()
        self._analysis_executor = futures.ThreadPoolExecutor(max_workers=1)

    def __str__(self):
        line = 51 * "-"
        n_res = len(self._analysis_results)
        status = self.status()
        ret = line
        ret += f"\nExperiment: {self.experiment_type}"
        ret += f"\nExperiment ID: {self.experiment_id}"
        if self._db_data.parent_id:
            ret += f"\nParent ID: {self._db_data.parent_id}"
        if self._child_data:
            ret += f"\nChild Experiment Data: {len(self._child_data)}"
        ret += f"\nStatus: {status}"
        if status == "ERROR":
            ret += "\n  "
            ret += "\n  ".join(self._errors)
        if self.backend:
            ret += f"\nBackend: {self.backend}"
        if self.tags:
            ret += f"\nTags: {self.tags}"
        ret += f"\nData: {len(self._result_data)}"
        ret += f"\nAnalysis Results: {n_res}"
        ret += f"\nFigures: {len(self._figures)}"
        return ret


@contextlib.contextmanager
def service_exception_to_warning():
    """Convert an exception raised by experiment service to a warning."""
    try:
        yield
    except Exception:  # pylint: disable=broad-except
        LOG.warning("Experiment service operation failed: %s", traceback.format_exc())


class ExperimentStatus(enum.Enum):
    """Class for experiment status enumerated type."""

    EMPTY = "experiment data is empty"
    INITIALIZING = "experiment jobs are being initialized"
    VALIDATING = "experiment jobs are validating"
    QUEUED = "experiment jobs are queued"
    RUNNING = "experiment jobs is actively running"
    CANCELLED = "experiment jobs or analysis has been cancelled"
    POST_PROCESSING = "experiment analysis is actively running"
    DONE = "experiment jobs and analysis have successfully run"
    ERROR = "experiment jobs or analysis incurred an error"

    def __json_encode__(self):
        return self.name

    @classmethod
    def __json_decode__(cls, value):
        return cls.__members__[value]  # pylint: disable=unsubscriptable-object


class AnalysisStatus(enum.Enum):
    """Class for analysis callback status enumerated type."""

    QUEUED = "analysis callback is queued"
    RUNNING = "analysis callback is actively running"
    CANCELLED = "analysis callback has been cancelled"
    DONE = "analysis callback has successfully run"
    ERROR = "analysis callback incurred an error"

    def __json_encode__(self):
        return self.name

    @classmethod
    def __json_decode__(cls, value):
        return cls.__members__[value]  # pylint: disable=unsubscriptable-object


@dataclasses.dataclass
class AnalysisCallback:
    """Dataclass for analysis callback status"""

    name: str = ""
    callback_id: str = ""
    status: AnalysisStatus = AnalysisStatus.QUEUED
    error_msg: Optional[str] = None
    event: Event = dataclasses.field(default_factory=Event)

    def __getstate__(self):
        # We need to remove the Event object from state when pickling
        # since events are not pickleable
        state = self.__dict__
        state["event"] = None
        return state

    def __json_encode__(self):
        return self.__getstate__()


def _series_to_service_result(
    series: pd.Series,
    service: IBMExperimentService,
    auto_save: bool,
    source: Optional[Dict[str, Any]] = None,
) -> AnalysisResult:
    """Helper function to convert dataframe to AnalysisResult payload for IBM experiment service.

    .. note::

        Now :class:`.AnalysisResult` is only used to save data in the experiment service.
        All local operations must be done with :class:`.AnalysisResultTable` dataframe.
        ExperimentData._analysis_results are totally decoupled from
        the model of IBM experiment service until this function is implicitly called.

    Args:
        series: Pandas dataframe Series (a row of dataframe).
        service: Experiment service.
        auto_save: Do auto save when entry value changes.

    Returns:
        Legacy AnalysisResult payload.
    """
    # TODO This must be done on experiment service rather than by client.
    qe_result = AnalysisResultData.from_table_element(**series.replace({np.nan: None}).to_dict())

    result_data = AnalysisResult.format_result_data(
        value=qe_result.value,
        extra=qe_result.extra,
        chisq=qe_result.chisq,
        source=source,
    )

    # Overwrite formatted result data dictionary with original objects.
    # The format_result_data method implicitly deep copies input value and extra field,
    # but it means the dictionary stores input objects with different object id.
    # This affects computation of error propagation with ufloats, because it
    # recognizes the value correlation with object id.
    # See test.curve_analysis.test_baseclass.TestCurveAnalysis.test_end_to_end_compute_new_entry.
    result_data["_value"] = qe_result.value
    result_data["_extra"] = qe_result.extra

    # IBM Experiment Service doesn't have data field for experiment and run time.
    # These are added to extra field so that these data can be saved.
    result_data["_extra"]["experiment"] = qe_result.experiment
    result_data["_extra"]["run_time"] = qe_result.run_time

    try:
        quality = ResultQuality(str(qe_result.quality).upper())
    except ValueError:
        quality = "unknown"

    experiment_service_payload = AnalysisResultDataclass(
        result_id=qe_result.result_id,
        experiment_id=qe_result.experiment_id,
        result_type=qe_result.name,
        result_data=result_data,
        device_components=list(map(to_component, qe_result.device_components)),
        quality=quality,
        tags=qe_result.tags,
        backend_name=qe_result.backend,
        creation_datetime=qe_result.created_time,
        chisq=qe_result.chisq,
    )

    service_result = AnalysisResult()
    service_result.set_data(experiment_service_payload)

    with contextlib.suppress(ExperimentDataError):
        service_result.service = service
        service_result.auto_save = auto_save

    return service_result


def _filter_analysis_results(
    search_key: Union[int, slice, str],
    data: pd.DataFrame,
) -> pd.DataFrame:
    """Helper function to search result data for given key.

    Args:
        search_key: Key to search for.
        data: Full result dataframe.

    Returns:
        Truncated dataframe.
    """
    out = _search_data(search_key, data)
    if isinstance(out, pd.Series):
        return pd.DataFrame([out])
    return out


@singledispatch
def _search_data(search_key, data):
    if search_key is None:
        return data
    raise TypeError(
        f"Invalid search key {search_key}. " f"This must be either int, slice or str type."
    )


@_search_data.register
def _search_with_int(
    search_key: int,
    data: pd.DataFrame,
):
    if search_key >= len(data):
        return None
    return data.iloc[search_key]


@_search_data.register
def _search_with_slice(
    search_key: slice,
    data: pd.DataFrame,
):
    out = data[search_key]
    if len(out) == 0:
        return None
    return out


@_search_data.register
def _search_with_str(
    search_key: str,
    data: pd.DataFrame,
):
    if search_key in data.index:
        # This key is table entry hash
        return data.loc[search_key]

    # This key is name of entry
    out = data[data["name"] == search_key]
    if len(out) == 0:
        return None
    return out