chore(blobby): penalize non-batched traffic for overflow detection

[xvello]

Problem

Sessions from very old posthog-js versions can still slow down a partition without triggering overflow, because they emit one message per event, instead of batching them together.

Batching was implemented in PostHog/posthog-js#694, released as v1.68.2 in 2023. Default batching settings are a linger of 2 seconds, or a size of 900kib

Changes

• OverflowManager has a new minPerCall property. Calls to observe consume at least this quantity
• The value is set to 1MB/session/batch on blobby (after messages in the same Kakfa batch are aggregated). This will make a session trigger overflow if it sends more than 5 batches per functional second after a 200 burst budget. Might not be too high? We'll see and increase if needed.

Does this work well for both Cloud and self-hosted?

How did you test this code?

[bretthoerner]

FAIL tests/main/ingestion-queues/session-recording/session-recordings-consumer.test.ts (8.895 s)
  ● ingester with consumeOverflow=false › when ingester.stop is called in teardown › overflow detection › should not trigger overflow if under threshold

Seems likely related?

[xvello]

new eslint rule

[bretthoerner]

Good idea!

I don't have a good fell for whether 1MB each is low or high, it sounds like you think it may not be high enough? If so that seems like a safer bet to start. 👍

[pauldambra]

👍

[posthog-bot]

This PR hasn't seen activity in a week! Should it be merged, closed, or further worked on? If you want to keep it open, post a comment or remove the stale label – otherwise this will be closed in another week.

[posthog-bot]

This PR was closed due to lack of activity. Feel free to reopen if it's still relevant.

[bretthoerner]

@xvello I assume you saw this, but just in case. Or maybe it's left unmerged on purpose?

[xvello]

Should be merged indeed, was not sure about the value to set though.