Measure compression, and precise sizes for manually emitted usecases (#97670)

This moves the compression step into the measured block, so we measure
the storage operation end to end including compression.

Also, this will now opt into precise metrics for the places where we
manually emit file sizes, like compressed nodestore, debug files and
bundles.

Author: Swatinem

src/sentry/models/debugfile.py

@@ -323,6 +323,7 @@ def create_dif_from_id(
         file.size,
         tags={"usecase": "debug-files", "compression": "none"},
         unit="byte",
+        precise=True,
     )
 
     dif = ProjectDebugFile.objects.create(

src/sentry/models/eventattachment.py

@@ -181,11 +181,9 @@ def putfile(cls, project_id: int, attachment: CachedAttachment) -> PutfileResult
             blob_path = "eventattachments/v1/" + FileBlob.generate_unique_path()
 
             storage = get_storage()
-            compressed_blob = zstandard.compress(data)
-
-            with measure_storage_operation(
-                "put", "attachments", size, len(compressed_blob), "zstd"
-            ):
+            with measure_storage_operation("put", "attachments", size) as metric_emitter:
+                compressed_blob = zstandard.compress(data)
+                metric_emitter.record_compressed_size(len(compressed_blob), "zstd")
                 storage.save(blob_path, BytesIO(compressed_blob))
 
         else:

src/sentry/profiles/task.py

@@ -1393,10 +1393,11 @@ def _process_vroomrs_transaction_profile(profile: Profile) -> bool:
             if prof.is_sampled():
                 with sentry_sdk.start_span(op="gcs.write", name="compress and write"):
                     storage = get_profiles_storage()
-                    compressed_profile = prof.compress()
                     with measure_storage_operation(
-                        "put", "profiling", len(json_profile), len(compressed_profile), "lz4"
-                    ):
+                        "put", "profiling", len(json_profile)
+                    ) as metric_emitter:
+                        compressed_profile = prof.compress()
+                        metric_emitter.record_compressed_size(len(compressed_profile), "lz4")
                         storage.save(prof.storage_path(), io.BytesIO(compressed_profile))
                 # we only run find_occurrences for sampled profiles, unsampled profiles
                 # are skipped
@@ -1456,10 +1457,11 @@ def _process_vroomrs_chunk_profile(profile: Profile) -> bool:
             chunk.normalize()
             with sentry_sdk.start_span(op="gcs.write", name="compress and write"):
                 storage = get_profiles_storage()
-                compressed_chunk = chunk.compress()
                 with measure_storage_operation(
-                    "put", "profiling", len(json_profile), len(compressed_chunk), "lz4"
-                ):
+                    "put", "profiling", len(json_profile)
+                ) as metric_emitter:
+                    compressed_chunk = chunk.compress()
+                    metric_emitter.record_compressed_size(len(compressed_chunk), "lz4")
                     storage.save(chunk.storage_path(), io.BytesIO(compressed_chunk))
             with sentry_sdk.start_span(op="processing", name="send chunk to kafka"):
                 payload = build_chunk_kafka_message(chunk)

src/sentry/tasks/assemble.py

@@ -522,6 +522,7 @@ def _create_or_update_artifact_bundle(
                 file.size,
                 tags={"usecase": "artifact-bundles", "compression": "none"},
                 unit="byte",
+                precise=True,
             )
 
             artifact_bundle = ArtifactBundle.objects.create(

src/sentry/utils/kvstore/bigtable.py

@@ -246,6 +246,7 @@ def _set(self, key: str, value: bytes, ttl: timedelta | None = None) -> None:
                 len(value),
                 tags={"usecase": "nodestore", "compression": self.compression},
                 unit="byte",
+                precise=True,
             )
 
         # Only need to write the column at all if any flags are enabled. And if