You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
New metrics identify which individual nodes are hot, ranked by mailbox backlog, message rate, and estimated wait. A low-cardinality node sitting on the critical path of a high-volume ingest throttles the whole stream; these metrics name that node by ID, so the bottleneck can be understood and the ingest query reworked.
The results card now shows what type each value is, so the number 42 and the string "42" no longer look identical, and a real null no longer looks like the word null.
Higher default outbound HTTP connection limits, so webhook outputs reach the throughput their parallelism setting asks for instead of being throttled below it.
Reading a node with a long history no longer costs memory proportional to that history, so long-lived nodes stay affordable to query.
Bugfixes:
A new snapshot-after-events setting lets a journaling deployment to control the number of journal events after the previous snapshot before creating a new snapshot, defaulting to 0 to snapshot on every sleep as before. Snapshots written by 2.1.x are migrated on read, but a data store this release has written cannot be read by 2.1.x, so downgrading after the upgrade is not possible. BREAKING-CHANGE
Nodes keep processing messages under clock skew. The max-catch-up-sleep setting is removed and must be dropped from any configuration that sets it. BREAKING-CHANGE
Server-Sent Events ingests no longer starve other requests to the same host, so an SSE stream can run alongside the rest of a deployment's HTTP traffic.