You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@ignite.apache.org by "Roman Puchkovskiy (Jira)" <ji...@apache.org> on 2022/10/31 10:58:00 UTC

[jira] [Created] (IGNITE-18018) Handle copy-on-write queue overflow in OutgoingSnapshot

Roman Puchkovskiy created IGNITE-18018:
------------------------------------------

             Summary: Handle copy-on-write queue overflow in OutgoingSnapshot
                 Key: IGNITE-18018
                 URL: https://issues.apache.org/jira/browse/IGNITE-18018
             Project: Ignite
          Issue Type: Improvement
          Components: persistence
            Reporter: Roman Puchkovskiy
             Fix For: 3.0.0-beta2


Currently, an OutgoingSnapshot does not restrict size of its internal queue of out-of-order rows to be sent to the receiver and does not provide any backpressure mechanism. This poses a threat of consuming too much memory (currently, it's heap memory) if the receiver is reading MV data slower than writes produce out-of-order data (or if the receiver has suddenly died and we still don't know about this).

One idea is to throttle writes, but this can only work fine if the created backpressure crosses the boundary created by the RAFT log.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)