Execute VerifyReplicationTasks as an individual activity #4656
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
yux0
reviewed
Jul 20, 2023
hehaifengcn
force-pushed
the
haifengh/v1.21.2-verify-test-parallel-pr
branch
from
fa6fdd6 to
e94efbb
Compare
wxing1292
reviewed
Jul 20, 2023
| @@ -374,41 +374,49 @@ func enqueueReplicationTasks(ctx workflow.Context, workflowExecutionsCh workflow | |||
| var lastActivityErr error | |||
|
|
|||
| for workflowExecutionsCh.Receive(ctx, &workflowExecutions) { | |||
There was a problem hiding this comment.
do you want to control the activity concurrency?
if so, here is a example: https://github.com/uber/cadence/blob/master/canary/concurrentExec.go
There was a problem hiding this comment.
explained offline. code already handle concurrency control.
wxing1292
approved these changes
Jul 20, 2023
hehaifengcn
force-pushed
the
haifengh/v1.21.2-verify-test-parallel-pr
branch
from
e94efbb to
990079d
Compare
meiliang86
approved these changes
Jul 21, 2023
yux0
approved these changes
Jul 21, 2023
wxing1292
pushed a commit
that referenced
this pull request
Jul 21, 2023
<!-- Describe what has changed in this PR --> **What changed?** Divide GenerateAndVerifyReplicationTasks activity into two activities: GenerateReplicationTasks (reuse previous one) and VerifyReplicationTasks <!-- Tell your future self why have you made these changes --> **Why?** Based on cluster tests, GenerateReplicationTasks is expensive (10ms latency for `GenerateLastHistoryReplicationTasks` call). In previous implementation, VerificationTasks runs after GenerateReplicationTasks and we only get ~60 RPS for GenerateAndVerifyReplicationTasks. By dividing the two, we can achieve ~100 RPS VerifyReplicationTasks for a single activity (bottleneck is still GenerateReplicationTasks because of 10ms latency). Also moved the special handling of WF not_found on target to VerifyReplicationTasks, which reduced # of `DescribeMutableState` call on source cluster. In previous implementation, `DescribeMutableState` is called for every replication task. Now we only call `DescribeMutableState` if WF was not found on target (which should be rare for steady state). The downside is that we can potentially replicate Zombie WF from source to target. But it should be avoidable by eliminating Zombie during migration process (i.e., delete WF on target if migration is incomplete). <!-- How have you verified this change? Tested locally? Added a unit test? Checked in staging env? --> **How did you test it?** Unit test & cluster tests. <!-- Assuming the worst case, what can be broken when deploying this change to production? --> **Potential risks** Low, the feature is disabled by default and only affect force replication workflow. <!-- Is this PR a hotfix candidate or require that a notification be sent to the broader community? (Yes/No) --> **Is hotfix candidate?** No.
hehaifengcn
added a commit
that referenced
this pull request
Jul 21, 2023
<!-- Describe what has changed in this PR --> **What changed?** Divide GenerateAndVerifyReplicationTasks activity into two activities: GenerateReplicationTasks (reuse previous one) and VerifyReplicationTasks <!-- Tell your future self why have you made these changes --> **Why?** Based on cluster tests, GenerateReplicationTasks is expensive (10ms latency for `GenerateLastHistoryReplicationTasks` call). In previous implementation, VerificationTasks runs after GenerateReplicationTasks and we only get ~60 RPS for GenerateAndVerifyReplicationTasks. By dividing the two, we can achieve ~100 RPS VerifyReplicationTasks for a single activity (bottleneck is still GenerateReplicationTasks because of 10ms latency). Also moved the special handling of WF not_found on target to VerifyReplicationTasks, which reduced # of `DescribeMutableState` call on source cluster. In previous implementation, `DescribeMutableState` is called for every replication task. Now we only call `DescribeMutableState` if WF was not found on target (which should be rare for steady state). The downside is that we can potentially replicate Zombie WF from source to target. But it should be avoidable by eliminating Zombie during migration process (i.e., delete WF on target if migration is incomplete). <!-- How have you verified this change? Tested locally? Added a unit test? Checked in staging env? --> **How did you test it?** Unit test & cluster tests. <!-- Assuming the worst case, what can be broken when deploying this change to production? --> **Potential risks** Low, the feature is disabled by default and only affect force replication workflow. <!-- Is this PR a hotfix candidate or require that a notification be sent to the broader community? (Yes/No) --> **Is hotfix candidate?** No.
hehaifengcn
added a commit
that referenced
this pull request
Jul 22, 2023
<!-- Describe what has changed in this PR --> **What changed?** Divide GenerateAndVerifyReplicationTasks activity into two activities: GenerateReplicationTasks (reuse previous one) and VerifyReplicationTasks <!-- Tell your future self why have you made these changes --> **Why?** Based on cluster tests, GenerateReplicationTasks is expensive (10ms latency for `GenerateLastHistoryReplicationTasks` call). In previous implementation, VerificationTasks runs after GenerateReplicationTasks and we only get ~60 RPS for GenerateAndVerifyReplicationTasks. By dividing the two, we can achieve ~100 RPS VerifyReplicationTasks for a single activity (bottleneck is still GenerateReplicationTasks because of 10ms latency). Also moved the special handling of WF not_found on target to VerifyReplicationTasks, which reduced # of `DescribeMutableState` call on source cluster. In previous implementation, `DescribeMutableState` is called for every replication task. Now we only call `DescribeMutableState` if WF was not found on target (which should be rare for steady state). The downside is that we can potentially replicate Zombie WF from source to target. But it should be avoidable by eliminating Zombie during migration process (i.e., delete WF on target if migration is incomplete). <!-- How have you verified this change? Tested locally? Added a unit test? Checked in staging env? --> **How did you test it?** Unit test & cluster tests. <!-- Assuming the worst case, what can be broken when deploying this change to production? --> **Potential risks** Low, the feature is disabled by default and only affect force replication workflow. <!-- Is this PR a hotfix candidate or require that a notification be sent to the broader community? (Yes/No) --> **Is hotfix candidate?** No.
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.
What changed?
Divide GenerateAndVerifyReplicationTasks activity into two activities: GenerateReplicationTasks (reuse previous one) and VerifyReplicationTasks
Why?
Based on cluster tests, GenerateReplicationTasks is expensive (10ms latency for
GenerateLastHistoryReplicationTaskscall). In previous implementation, VerificationTasks runs after GenerateReplicationTasks and we only get ~60 RPS for GenerateAndVerifyReplicationTasks. By dividing the two, we can achieve ~100 RPS VerifyReplicationTasks for a single activity (bottleneck is still GenerateReplicationTasks because of 10ms latency).Also moved the special handling of WF not_found on target to VerifyReplicationTasks, which reduced # of
DescribeMutableStatecall on source cluster. In previous implementation,DescribeMutableStateis called for every replication task. Now we only callDescribeMutableStateif WF was not found on target (which should be rare for steady state). The downside is that we can potentially replicate Zombie WF from source to target. But it should be avoidable by eliminating Zombie during migration process (i.e., delete WF on target if migration is incomplete).How did you test it?
Unit test & cluster tests.
Potential risks
Low, the feature is disabled by default and only affect force replication workflow.
Is hotfix candidate?
No.