[ML] Improve resuming a data frame analytics job stopped during inference #67623
Conversation
This file contains 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
If a DFA job is stopped while in the inference phase, after resuming we should start inference immediately. However, this is currently not the case. Inference is tied in `AnalyticsProcessManager` and thus we start a process, load data, restore state, etc., until we get to start inference. This commit gets rid of this unnecessary delay by factoring inference out as an independent step and ensuring we can resume straight from that phase upon restarting a job.
|
Pinging @elastic/ml-core (:ml) |
przemekwitek
approved these changes
Jan 18, 2021
| * Updates the progress tracker with potentially new in-between phases | ||
| * that were introduced in a later version while making sure progress indicators | ||
| * are correct. | ||
| * @param analysisPhases the new analysis phases |
There was a problem hiding this comment.
To me the phrase "new analysis phases" suggests that analysisPhases is a diff compared to some previous state whereas IIUC this is the full new list of phases.
Do you think it could be rephrased?
|
@elasticmachine update branch |
dimitris-athanasiou
deleted the
improve-resume-dfa-job-in-inference-phase
branch
dimitris-athanasiou
added a commit
to dimitris-athanasiou/elasticsearch
that referenced
this pull request
Jan 18, 2021
…c#67623) If a DFA job is stopped while in the inference phase, after resuming we should start inference immediately. However, this is currently not the case. Inference is tied in `AnalyticsProcessManager` and thus we start a process, load data, restore state, etc., until we get to start inference. This commit gets rid of this unnecessary delay by factoring inference out as an independent step and ensuring we can resume straight from that phase upon restarting a job. Backport of elastic#67623
dimitris-athanasiou
added a commit
that referenced
this pull request
Jan 18, 2021
#67669) If a DFA job is stopped while in the inference phase, after resuming we should start inference immediately. However, this is currently not the case. Inference is tied in `AnalyticsProcessManager` and thus we start a process, load data, restore state, etc., until we get to start inference. This commit gets rid of this unnecessary delay by factoring inference out as an independent step and ensuring we can resume straight from that phase upon restarting a job. Backport of #67623
droberts195
changed the title
dimitris-athanasiou
added a commit
to dimitris-athanasiou/elasticsearch
that referenced
this pull request
Jan 19, 2021
In elastic#67623 I moved persisting the data counts at the end of a data frame analytics job into a `FinalStep` class. However, I forgot to execute the index request with ML origin resulting in authentication problems if the user that runs the DFA job does not have read privileges in the ML stats index. This commit fixes this by executing that index request with ML origin.
dimitris-athanasiou
added a commit
that referenced
this pull request
Jan 19, 2021
In #67623 I moved persisting the data counts at the end of a data frame analytics job into a `FinalStep` class. However, I forgot to execute the index request with ML origin resulting in authentication problems if the user that runs the DFA job does not have read privileges in the ML stats index. This commit fixes this by executing that index request with ML origin.
dimitris-athanasiou
added a commit
that referenced
this pull request
Jan 19, 2021
… (#67683) In #67623 I moved persisting the data counts at the end of a data frame analytics job into a `FinalStep` class. However, I forgot to execute the index request with ML origin resulting in authentication problems if the user that runs the DFA job does not have read privileges in the ML stats index. This commit fixes this by executing that index request with ML origin. Backport of #67674
dimitris-athanasiou
added a commit
to dimitris-athanasiou/elasticsearch
that referenced
this pull request
Jan 25, 2021
Now that data frame analytics jobs can be resumed straight into the inference phase, we need to ensure data counts are persisted at the end of the analysis step and restored when the job is started again. This commit removes the need for storing the progress on start as a task parameter. Instead, when the task gets assigned we now restore all stats by making a call to the get stats API. Additionally, we now ensure that an allocated task that hasn't had its `StatsHolder` restored yet is treated as a stopped task from the get stats API, which means we will report the stored stats. Relates elastic#67623
dimitris-athanasiou
added a commit
that referenced
this pull request
Jan 26, 2021
Now that data frame analytics jobs can be resumed straight into the inference phase, we need to ensure data counts are persisted at the end of the analysis step and restored when the job is started again. This commit removes the need for storing the progress on start as a task parameter. Instead, when the task gets assigned we now restore all stats by making a call to the get stats API. Additionally, we now ensure that an allocated task that hasn't had its `StatsHolder` restored yet is treated as a stopped task from the get stats API, which means we will report the stored stats. Relates #67623
dimitris-athanasiou
added a commit
that referenced
this pull request
Jan 26, 2021
… (#67979) Now that data frame analytics jobs can be resumed straight into the inference phase, we need to ensure data counts are persisted at the end of the analysis step and restored when the job is started again. This commit removes the need for storing the progress on start as a task parameter. Instead, when the task gets assigned we now restore all stats by making a call to the get stats API. Additionally, we now ensure that an allocated task that hasn't had its `StatsHolder` restored yet is treated as a stopped task from the get stats API, which means we will report the stored stats. Relates #67623 Backport of #67937
34 tasks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
If a DFA job is stopped while in the inference phase, after
resuming we should start inference immediately. However, this
is currently not the case. Inference is tied in
AnalyticsProcessManagerand thus we start a process, load data, restore state, etc., until
we get to start inference.
This commit gets rid of this unnecessary delay by factoring inference
out as an independent step and ensuring we can resume straight from
that phase upon restarting a job.