[DB-26-19] Warn when projection state becomes greater than 8mb

[lakshdeepsingheventstore]

Added : A warning to the log when a projection state size becomes greater than 8 MB. https://eventstore.aha.io/develop/requirements/DB-26-19

[timothycoleman]

• in the pr description, this isn't a bug so we don't need to put fixed. we can use Added:
• im not sure, but does this apply to projection state specifically or all emitted projection events?

[timothycoleman]

^

hayley is back :)

[hayley-jean]

Thanks! There's one more place where the state size needs to be checked

What you have here works for the projection result that will be written to the $projection-{projection_name}-result stream. This result is only written for Queries, or for Continuous/OneTime projections that call outputState() in the projection definition.

The size of the state that makes up the checkpoints written to the $projection-{projection_name}-checkpoint stream and $projection-{projection_name}-{partition_name}-checkpoint streams also needs to be checked.

Probably the best place to check and log a warning for the state size is when StateUpdated is called in the CoreProjectionCheckpointManager (in addition to the check you're doing here in ResultEventEmitter)

[lakshdeepsingheventstore]

The size of the state that makes up the checkpoints written to the $projection-{projection_name}-checkpoint stream and $projection-{projection_name}-{partition_name}-checkpoint streams also needs to be checked.

Making changes to CoreProjectionCheckpointManager solves the same purpose as was being done by the changes in ResultEventEmitter file. So in the latest commit/changes, I have removed the changes from ResultEventEmitter, and added those in CoreProjectionCheckpointManager in the method StateUpdated (this would make it more clear that the state size is being checked)
However, to exclusively check for state sizes for the checkpoint streams, separate checks had to be added to CoreProjectionCheckpointWriter and PartitionStateUpdateManager files, since the check in CoreProjectionCheckpointManager does not log anything when the projection Checkpoint is written.

[hayley-jean]

Sorry about not including this in my earlier review, but I learned new things about how the result of a projection interacts with the state while testing this. Basically, we need to include both the state and result in PartitionState when checking the size of the projection state.

I previously thought that the result was only written if you had outputState() defined on your projection, but it turns out that the check for that is commented out.

Instead, a result is always written for continuous projections (just not always in a result stream). If the projection includes a transformBy() function then the result will be the result of that transformation run on the state. Otherwise it simply duplicates the state into the result.

The result is included with the state when the checkpoint is written, so this can effectively double the state size that is written to disk.

I think the best way to handle this would be to have a Size property on PartitionState that returns something like _state.Length + _result?.Length ?? 0; and use that for the warning log. Otherwise, by the time the warning is logged at 8mb, you are already potentially over the 16mb limit.

Let me know if you want a sample projection to test this case