Enable bfloat16/directIO on int quantized vector types #137441
Draft
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
thecoop
force-pushed
the
int-hnsw-bfloat16-enabled
branch
from
fc22aa4 to
30fb01b
Compare
thecoop
force-pushed
the
int-hnsw-bfloat16-enabled
branch
from
30fb01b to
2cf09b7
Compare
|
Using the new quantizer has changed query behaviour, requiring some investigation |
…astic#137023) Following elastic#136813, we expose to data stream lifecycle the new sampling method config in the downsampling API. This will allow users to configure the sampling method directly in the lifecycle configuration. For example: ``` PUT _data_stream/my-ds/_lifecycle { "data_retention": "10d", "downsampling_method": "last_value", "downsampling": [ { "after": "1d", "fixed_interval": "5m } ] } ```
As of elastic#133954, we clone indices before performing the force-merge step in the `searchable_snapshot` action. On slow CI servers, 10 seconds for the index to go through the whole `searchable_snapshot` action isn't enough, so we bump the timeout to 20 seconds. I looked at the logs of a few test failures, and ILM was clearly still progressing when the test timed out. I didn't identify any particular step that was taking extraordinarily long; there were always just a few steps that took a bit longer. I would love to make these tests faster rather than bumping the timeout, but the `searchable_snapshot` action is simply one of the largest ILM actions and ILM itself isn't particularly fast. That being said, if a timeout of 20 seconds proves to be insufficient (i.e. test failures come back), I do think it's worth having a look at reducing the runtime of the tests somehow first before we increase the timeout further. Closes elastic#137149 Closes elastic#137151 Closes elastic#137152 Closes elastic#137153 Closes elastic#137156 Closes elastic#137166 Closes elastic#137167 Closes elastic#137192
…elastic#137520) This reverts commit 4d3e27f. And then re-applies the revert to elastic#134738, so we only get an effective revert of elastic#134511.
When Dataframes are disabled, do not try to call the Action and instead assume there are no trained models associated with a dataframe job. This avoids a transport exception for the disabled action.
…LMIntegration {targetCluster=FOLLOWER} elastic#137494
…hRollover {targetCluster=FOLLOWER} elastic#137528
This adds a new abstract method to MapperTestCase for checking whether a particular field type configuration supports competitive iterators during sorting.
…hat it was resolved to (elastic#136632) * alexey-ivanov-es work * iter * field caps: local resolution working and tested * CCS impl * using CPS SAF resolution if present * some TODOs to check later on * [CI] Update transport version definitions * iter * iter * iter * iter * Update docs/changelog/136632.yaml * [CI] Auto commit changes from spotless * iter * iter * iter * iter * only return resolved to if minTransportVersion supports it * iter * updated comment based on feedback --------- Co-authored-by: elasticsearchmachine <[email protected]>
…lastic#137404) In elastic#137139, ES819DocValuesFormat version will increment from 0 to 1. For bwc tests between version 0 and version 1, we need a copy of version 0. So copy the existing ES819TSDBDocValuesConsumer to ES819TSDBDocValuesConsumerVersion0 and add TestES819TSDBDocValuesFormatVersion0 which uses this consumer. Since ES819TSDBDocValuesConsumer and ES819TSDBDocValuesConsumerVersion0 are currently identical, this does not test anything yet, but will once elastic#137139 increments the version.
Mainly this is about removing the trappy implicit default master node timeout in resize (shrink/split/clone) requests as per elastic#107984, but also we drop the unnecessary `ActionType` subclass and the test-only `ResizeRequestBuilder`, migrating the affected tests over to a new `ResizeIndexTestUtils` instead. We also fix the order of index-name arguments and make the `ResizeType` a mandatory parameter rather than (trappily) defaulting to `SHRINK`.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
15 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.