Add byte-size limit to projection query buffer to prevent OOM on large events

[JustinPihony]

Summary

The akka.persistence.r2dbc.query.buffer-size setting controls how many events are buffered in memory during projection replay, but it is purely count-based (default: 1000). When individual events are large (e.g. several hundred KB each), a buffer of 1000 events can consume hundreds of MB or more, causing OOM errors.

Problem

When a view projection does a full replay over a large number of events with large payloads, the count-based buffer can exhaust available heap. The workaround is to manually reduce buffer-size, but this requires knowing the average event size ahead of time and is fragile — the right value differs per service and per deployment.

Proposed solution

Add an optional byte-size cap to the query buffer so that the buffer stops filling before it exceeds a configured memory threshold, regardless of event count:

akka.persistence.r2dbc.query {
  buffer-size = 1000          # existing: max number of events
  buffer-max-bytes = 64 MiB   # new: optional max total payload bytes in buffer
}

When buffer-max-bytes is set, the buffer would stop accepting new entries once the cumulative serialized payload size exceeds the limit, even if the count has not yet reached buffer-size. This provides a memory-sensitive upper bound that automatically adapts to event sizes without per-service tuning.

Context

• Projection buffer config lives in reference.conf under akka.persistence.r2dbc.query.buffer-size
• Event replay uses this same buffer
• The same pattern (large-event OOM on projection replay) has been observed across multiple independent deployments