[
https://issues.apache.org/jira/browse/IGNITE-23908?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Roman Puchkovskiy updated IGNITE-23908:
---------------------------------------
Description:
In IGNITE-23550, it was discovered that fsyncs we do when writing each revision
checksum (that is, on each Metastorage revision) are too costly. They were made
durable to be durable to solve the 'durable side effects for a rolled-back
Metastorage updated' described in IGNITE-23910, but the price of the current
design is too high.
In IGNITE-23910, we'll work in another solution to the same problem. For now
the best thing we can do is just turn off the fsyncing as cases like
IGNITE-23910 are extremely rare and we should not pay with everyday performance
for them.
> Do not fsync Metastorage revision checksum writes
> -------------------------------------------------
>
> Key: IGNITE-23908
> URL: https://issues.apache.org/jira/browse/IGNITE-23908
> Project: Ignite
> Issue Type: Improvement
> Reporter: Roman Puchkovskiy
> Assignee: Roman Puchkovskiy
> Priority: Major
> Time Spent: 10m
> Remaining Estimate: 0h
>
> In IGNITE-23550, it was discovered that fsyncs we do when writing each
> revision checksum (that is, on each Metastorage revision) are too costly.
> They were made durable to be durable to solve the 'durable side effects for a
> rolled-back Metastorage updated' described in IGNITE-23910, but the price of
> the current design is too high.
> In IGNITE-23910, we'll work in another solution to the same problem. For now
> the best thing we can do is just turn off the fsyncing as cases like
> IGNITE-23910 are extremely rare and we should not pay with everyday
> performance for them.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
