Repository navigation
ref(reprocessing): Optionally pass unprocessed copy inline instead of via redis - #126810
Conversation
|
Is there any linear ticket context to this? |
| from_symbolicate=from_symbolicate, | ||
| has_attachments=has_attachments, | ||
| data=data, | ||
| unprocessed=unprocessed, |
There was a problem hiding this comment.
Is all the data in this dict JSON encodable? Tasks will fail if there are any non-json encodable values in the object.
There was a problem hiding this comment.
Yes, this is encodable. It contains the original data.
| has_attachments=has_attachments, | ||
| symbolicate_functions=symbolicate_functions, | ||
| data=data, | ||
| unprocessed=unprocessed, |
There was a problem hiding this comment.
Should unprocessed and cache_key be mutually exclusive? I would think that the end state is that we don't have both being used as the cache key is redundant when we also have the unprocessed + data parameters present.
There was a problem hiding this comment.
In the end state they will be mutually exclusive and eventually cache_key will be removed from the tasks. During rollout, we also double-write for some time so we can safely roll back at any point.
607e45e to
dd83958
Compare
Removing the Redis processing store requires another way to preserve the original event payload for reprocessing.
With this change, the unprocessed copy travels inline through symbolication, processing, and save tasks into Nodestore. Redis remains the fallback when no inline copy is supplied.
Keeping
store.disable-processing-store=falseduring deployment preserves compatibility with older workers in any rollout order. Redis backups can be disabled after all workers are updated.