Thomas Mueller created OAK-11766:
------------------------------------
Summary: Write Throttling Mechanism
Key: OAK-11766
URL: https://issues.apache.org/jira/browse/OAK-11766
Project: Jackrabbit Oak
Issue Type: Improvement
Reporter: Thomas Mueller
Asynchronous index updates can fall behind writes in the following cases:
* When using the document node store (it is theoretically also possible with
other node stores, but I didn't see the behavior with others), and
* if many threads concurrently write to the repository (eg. using POST
servlets), and
* if multiple cluster nodes concurrently write, and
* if there are many indexes that are relatively expensive.
The challenge is that there is only one thread, per async lane, that can update
the indexes. What I have seen is that this indexing lane can fall behind, and
possibly (if write load persists for multiple hours) this indexing lane never
recovers within a reasonable time.
To resolve the problem, I propose that we introduce a delay mechanism to
configure delays in Session.save() calls.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
