Update sequence impl to tune controller memory consumption #2387
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
rabbah
added
controller
review
rabbah
force-pushed
the
sequence_avoid_keeping_N_in_mem
branch
2 times, most recently
from
828162f
to
c061fe0
Compare
|
@tardieu @markusthoemmes @ioana-blue might any of you want to review this? |
rabbah
force-pushed
the
sequence_avoid_keeping_N_in_mem
branch
from
c061fe0
to
792d3f2
Compare
|
@starpit as I noted on slack, we should have unit tests for the new/refactored code. It's far more testable now and we can get good coverage. If you don't have bandwidth for this, I will work on that tomorrow. |
rabbah
force-pushed
the
sequence_avoid_keeping_N_in_mem
branch
from
792d3f2
to
dd99793
Compare
…gs in heap. switch to scheduleOnce+weakrefs for timeout handling in SequenceActions switch SequenceAccounting to store array of ActivationId rather than array of String -- cheaper in memory use better (non-dragging) impl of withTimeout use a getAndSet(null) pattern to avoid two copies of responses being alive simultaneously
Remove pre-allocated array and use mutable array buffer. Terminate the fold when shortcuirting. Use copy constructor for fail. Remove promise. Dusting of tests.
rabbah
force-pushed
the
sequence_avoid_keeping_N_in_mem
branch
4 times, most recently
from
239b6b9
to
62d65d1
Compare
rabbah
force-pushed
the
sequence_avoid_keeping_N_in_mem
branch
from
62d65d1
to
8c2175c
Compare
rabbah
force-pushed
the
sequence_avoid_keeping_N_in_mem
branch
from
8a41fdd
to
f97f892
Compare
| .map(currentMax => Some(Math.max(prevMax, currentMax))) | ||
| .getOrElse(prevMemoryLimit) | ||
| }.getOrElse(newMemoryLimit) | ||
| } |
There was a problem hiding this comment.
Is this the same as:
(prevMemoryLimit ++ newMemoryLimit).reduceOption(Math.max)
?
| protected[actions] case class SequenceAccounting( | ||
| atomicActionCnt: Int, | ||
| previousResponse: AtomicReference[ActivationResponse], | ||
| logs: mutable.ArrayBuffer[ActivationId], |
There was a problem hiding this comment.
Do we rely on this being an ArrayBuffer anywhere, or could it aswell be the more general Buffer?
| /** The previous activation failed (this is used when there is no activation record or an internal error. */ | ||
| def fail(failureResponse: ActivationResponse, activationId: Option[ActivationId]) = { | ||
| require(!failureResponse.isSuccess) | ||
| activationId.foreach(logs += _) |
There was a problem hiding this comment.
logs.appendAll(activationId)?
rabbah
approved these changes
Jun 20, 2017
houshengbo
pushed a commit
to houshengbo/openwhisk
that referenced
this pull request
Jun 23, 2017
- switch to scheduleOnce+weakrefs for timeout handling in SequenceActions - switch SequenceAccounting to store array of ActivationId rather than array of String -- cheaper in memory - use better (non-dragging) impl of withTimeout - use a getAndSet(null) pattern to avoid two copies of responses being alive simultaneously - refactor top level sequence scheduler to eliminate promises
houshengbo
pushed a commit
to houshengbo/openwhisk
that referenced
this pull request
Jun 23, 2017
- switch to scheduleOnce+weakrefs for timeout handling in SequenceActions - switch SequenceAccounting to store array of ActivationId rather than array of String -- cheaper in memory - use better (non-dragging) impl of withTimeout - use a getAndSet(null) pattern to avoid two copies of responses being alive simultaneously - refactor top level sequence scheduler to eliminate promises
jonpspri
pushed a commit
to jonpspri/openwhisk
that referenced
this pull request
Jul 4, 2017
- switch to scheduleOnce+weakrefs for timeout handling in SequenceActions - switch SequenceAccounting to store array of ActivationId rather than array of String -- cheaper in memory - use better (non-dragging) impl of withTimeout - use a getAndSet(null) pattern to avoid two copies of responses being alive simultaneously - refactor top level sequence scheduler to eliminate promises
houshengbo
pushed a commit
to houshengbo/openwhisk
that referenced
this pull request
Jul 17, 2017
- switch to scheduleOnce+weakrefs for timeout handling in SequenceActions - switch SequenceAccounting to store array of ActivationId rather than array of String -- cheaper in memory - use better (non-dragging) impl of withTimeout - use a getAndSet(null) pattern to avoid two copies of responses being alive simultaneously - refactor top level sequence scheduler to eliminate promises
houshengbo
pushed a commit
to houshengbo/openwhisk
that referenced
this pull request
Jul 21, 2017
- switch to scheduleOnce+weakrefs for timeout handling in SequenceActions - switch SequenceAccounting to store array of ActivationId rather than array of String -- cheaper in memory - use better (non-dragging) impl of withTimeout - use a getAndSet(null) pattern to avoid two copies of responses being alive simultaneously - refactor top level sequence scheduler to eliminate promises
houshengbo
pushed a commit
to houshengbo/openwhisk
that referenced
this pull request
Nov 13, 2017
- switch to scheduleOnce+weakrefs for timeout handling in SequenceActions - switch SequenceAccounting to store array of ActivationId rather than array of String -- cheaper in memory - use better (non-dragging) impl of withTimeout - use a getAndSet(null) pattern to avoid two copies of responses being alive simultaneously - refactor top level sequence scheduler to eliminate promises
houshengbo
pushed a commit
to houshengbo/openwhisk
that referenced
this pull request
Nov 13, 2017
- switch to scheduleOnce+weakrefs for timeout handling in SequenceActions - switch SequenceAccounting to store array of ActivationId rather than array of String -- cheaper in memory - use better (non-dragging) impl of withTimeout - use a getAndSet(null) pattern to avoid two copies of responses being alive simultaneously - refactor top level sequence scheduler to eliminate promises
houshengbo
pushed a commit
to houshengbo/openwhisk
that referenced
this pull request
Nov 13, 2017
- switch to scheduleOnce+weakrefs for timeout handling in SequenceActions - switch SequenceAccounting to store array of ActivationId rather than array of String -- cheaper in memory - use better (non-dragging) impl of withTimeout - use a getAndSet(null) pattern to avoid two copies of responses being alive simultaneously - refactor top level sequence scheduler to eliminate promises
houshengbo
pushed a commit
to houshengbo/openwhisk
that referenced
this pull request
Nov 13, 2017
- switch to scheduleOnce+weakrefs for timeout handling in SequenceActions - switch SequenceAccounting to store array of ActivationId rather than array of String -- cheaper in memory - use better (non-dragging) impl of withTimeout - use a getAndSet(null) pattern to avoid two copies of responses being alive simultaneously - refactor top level sequence scheduler to eliminate promises
houshengbo
pushed a commit
to houshengbo/openwhisk
that referenced
this pull request
Nov 14, 2017
- switch to scheduleOnce+weakrefs for timeout handling in SequenceActions - switch SequenceAccounting to store array of ActivationId rather than array of String -- cheaper in memory - use better (non-dragging) impl of withTimeout - use a getAndSet(null) pattern to avoid two copies of responses being alive simultaneously - refactor top level sequence scheduler to eliminate promises
houshengbo
pushed a commit
to houshengbo/openwhisk
that referenced
this pull request
Nov 14, 2017
- switch to scheduleOnce+weakrefs for timeout handling in SequenceActions - switch SequenceAccounting to store array of ActivationId rather than array of String -- cheaper in memory - use better (non-dragging) impl of withTimeout - use a getAndSet(null) pattern to avoid two copies of responses being alive simultaneously - refactor top level sequence scheduler to eliminate promises
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.
Fixes #2386
This PR changes SequenceActions to use foldLeft rather than scanLeft. The change requires the introduction of a new accumulator data structure, to maintain the accumulated state for the reduction. The accumulated state includes an array of handles to logs (for the components), the cumulative duration, and a few other bits needed for creation of the final sequence activation record.
This fix also patches a closely related memory drag, caused by the way timeouts of sequence invocations are handled. See #2294 which patches other parts of the controller for the same kind of timeout drag. This patch needs to be a little different, due to the complex way the SequenceActions impl uses nested Promises.
PG3 625 🔵