[CHASM] Support Non-Workflow Mutable State: Part 1 #7595
Conversation
|
|
||
| // TODO: we may need to handle the case where mutable state only starts to generate events during middle of its execution. | ||
| // In that case targetVersionHistories maybe empty and sourceVersionHistories is not empty. | ||
| // The LCA logic doesn't work in that case today. |
There was a problem hiding this comment.
can we use transition history in this case?
There was a problem hiding this comment.
no quite following. This method is specifically for dealing with events, and see which events are missing on the target side. Transition history doesn't give any info re. events.
There was a problem hiding this comment.
Distinguish this case seems straight forward:
target transition history is not empty but target version history is empty?
| if err != nil { | ||
| return 0, err | ||
|
|
||
| if !versionhistory.IsEmptyVersionHistory(versionHistory) { |
There was a problem hiding this comment.
Use IsCurrentVersionHistoryEmpty?
There was a problem hiding this comment.
We still need to get the current version history (then to get the first item) if it's not empty.
| } | ||
| } | ||
|
|
||
| return common.EmptyVersion, nil |
There was a problem hiding this comment.
is this expected in some case to return empty version?
There was a problem hiding this comment.
I feel we have bunch of logic probably trying to handle backward compatibility (by checking for version history is nil) and return EmptyVersion. Like GetCurrentVersion(). If you have any context on that, plz let me know.
I'd say for any new workflows/chasm runs, it's not expected to reach this line (EmptyVersion can still be returned above for local namespace). For very old workflows, i am not sure.
| if ms.IsWorkflowExecutionRunning() { | ||
| // Do NOT use ms.IsWorkflowExecutionRunning() for the check. | ||
| // Zombie workflow is not considered running but also not closed. | ||
| if ms.executionState.State != enumsspb.WORKFLOW_EXECUTION_STATE_COMPLETED { |
There was a problem hiding this comment.
I need to double check all the callers. This is breaking ndc workflow right now, which may call it when workflow is in zombie state.
There was a problem hiding this comment.
Guaranteeing caller not calling this when workflow is zombie will require a big refactoring. For now, I special handled zombie case and returns LastWriteVersion instead.
| ) { | ||
| return historyi.TransactionPolicyActive, serviceerror.NewInternal("Workflow cannot suppress workflow by older workflow") | ||
| } | ||
|
|
||
| // if workflow is in zombie or finished state, keep as is |
There was a problem hiding this comment.
Move those lines seems changed semantic a little bit, what's the reason of moving?
|
|
||
| // TODO: we may need to handle the case where mutable state only starts to generate events during middle of its execution. | ||
| // In that case targetVersionHistories maybe empty and sourceVersionHistories is not empty. | ||
| // The LCA logic doesn't work in that case today. |
There was a problem hiding this comment.
Distinguish this case seems straight forward:
target transition history is not empty but target version history is empty?
| transactionPolicy historyi.TransactionPolicy, | ||
| workflowEventsSeq []*persistence.WorkflowEvents, | ||
| ) { | ||
| if transactionPolicy != historyi.TransactionPolicyActive { |
There was a problem hiding this comment.
Are we replicating the vector clock from active to passive?
What changed?
Changes are mainly on replication side:
Why?
How did you test it?
Potential risks
Documentation
Is hotfix candidate?