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
A legacy discovery set written without a TTL no longer keeps retired queues and job classes listed forever. Versions before v2.0 wrote {prefix}:discovery:queues and {prefix}:discovery:jobs with no expiry, and v3.5.0 no longer writes those keys, so on an install that came from v1.x every member stayed in listQueues() and listJobs(), which is what v3.5.0 set out to fix. A reader now gives such a key the aggregated TTL the first time it reads it. A legacy key that already has a TTL is left alone.
Changed
queue-metrics:calculate reads the discovered job list once per run instead of once per queue. On Redis that saves about four round trips for every queue after the first.
RedisMetricsStore::expireIfPersistent() sets a TTL on a key only if it exists and has none.
Documentation
The discovery page says that reading discovery prunes it, so the storage connection must be writable (not a read-only replica). It also says that only a job starting marks a queue as seen, and that a scheduler host still on v3.4 does not see entries first seen on v3.5 workers during a rolling deploy.