[
https://issues.apache.org/jira/browse/IGNITE-15129?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Alexey Scherbakov updated IGNITE-15129:
---------------------------------------
Description:
Current timestamp implementation doesn't include globalId part, so is useful
only then all transactions ids are generated on single node.
It should be improved by adding global unique jvm id to the timestamp.
Additionally, current timestamp uses nanoSeconds, which is far away from real
time.
Instead, NTP timestamp can be used as a localTime part.
IgniteUuid can be used for this purpose as well - it's the simplest solution.
Also, the performance of current implementation may not be optimal - probably
it should be CAS powered or/and use optimistic locking (StampedLock).
was:
Current timestamp implementation doesn't include globalId part, so is useful
only then all transactions ids are generated on single node.
It should be improved by adding global unique jvm id to the timestamp.
Additionally, current timestamp uses nanoSeconds, which is far away from real
time.
Instead, NTP timestamp can be used as a localTime part.
IgniteUuid can be used for this purpose as well - it's the simplest solution.
Also, the perfornace of current implementation may not be optimal - probably it
should be CAS powered or/and use optimistic locking (StampedLock).
> Improve timestamp generation for transactions
> ---------------------------------------------
>
> Key: IGNITE-15129
> URL: https://issues.apache.org/jira/browse/IGNITE-15129
> Project: Ignite
> Issue Type: Improvement
> Reporter: Alexey Scherbakov
> Priority: Major
> Labels: iep-61, ignite-3
> Fix For: 3.0.0-alpha3
>
>
> Current timestamp implementation doesn't include globalId part, so is useful
> only then all transactions ids are generated on single node.
> It should be improved by adding global unique jvm id to the timestamp.
> Additionally, current timestamp uses nanoSeconds, which is far away from real
> time.
> Instead, NTP timestamp can be used as a localTime part.
> IgniteUuid can be used for this purpose as well - it's the simplest solution.
> Also, the performance of current implementation may not be optimal - probably
> it should be CAS powered or/and use optimistic locking (StampedLock).
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
