82017dd301
Some checks failed
deploy / deploy (push) Failing after 47s
Proof: Live investigation showed doran-1 entered NodeNotReady with kubelet SystemOOM and TLS/control-plane timeouts while near-intents-ingest, history-writer, and operator-dashboard were the largest Node memory consumers. This commit adds websocket publish backpressure for the raw quote firehose and pod memory guardrails for the affected services. Assumptions: Dropping quote frames while Kafka publishing is backpressured is safer than allowing unbounded in-flight publishes to take down the single-node cluster; retained Kafka/Postgres history remains best-effort under overload until the platform has enough capacity for full raw retention. Still fake: This does not add durable queue spillover for skipped raw websocket frames, does not resize the node, and does not prove fee-complete trading PnL. |
||
|---|---|---|
| .. | ||
| k8s/base | ||
| redpanda | ||