You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@pulsar.apache.org by Apache Pulsar Slack <ap...@gmail.com> on 2020/07/28 09:11:04 UTC
Slack digest for #dev - 2020-07-28
2020-07-27 11:12:28 UTC - Ankur Jain: @Ankur Jain has joined the channel
----
2020-07-27 11:27:45 UTC - Ankur Jain: @Matteo Merli Is it possible for a replicated sub to not find relevant snapshot in the cache consistently such it will never be able to publish `ReplicatedSubscriptionsUpdate` ? Consider a topic being produced to at consistent rate (100/second), so 1 snapshot (interval being 1 second) pertains to roughly 100 messages. If max snaps in `ReplicatedSubscriptionSnapshotCache` = 10, then this broker-side cache for a subscription would be able to resolve snapshots (using `advancedMarkDeletePosition`) for latest ~1000 (10 x 100) messages that were dispatched from the broker. Is this understanding correct?
Building upon the above understanding, let's say a consumer has ~5000 messages in flight at any point of time and it keeps sending acks for the earliest of unacked messages. A relevant snapshot for such messages will not be found. As and when consumer ack advances, read messages by the consumers are also advancing and snapshot cache is deleting old snapshots and hence `ReplicatedSubscriptionsUpdate` never gets written to topic. Can this happen when `unacked messages > cache_count*snap_itvl*topic_publish_rate` ?
----
2020-07-27 22:42:31 UTC - mal: @mal has joined the channel
----
2020-07-28 05:42:30 UTC - Sean Moir: @Sean Moir has joined the channel
----