[
https://issues.apache.org/jira/browse/SENTRY-1767?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16084788#comment-16084788
]
Na Li commented on SENTRY-1767:
-------------------------------
[~akolb] I have not fully figure out how to achieve that.
One way is to have a new table in DB whose entry contains
[CurrentAuthzPathsSnapshotID, time it starts, sentry server ID or name]. Once
full snapshot is done and persisted, the CurrentAuthzPathsSnapshotID increases,
and that entry does not take effect and won't stop another sentry server from
getting a full snapshot.
When a sentry server wants to get full snapshot, if such entry exist for
CurrentAuthzPathsSnapshotID, not itself, and not old enough ("current time" -
"time it starts"), it will skip and wait for next run.
> Multiple followers should not create full snapshot
> --------------------------------------------------
>
> Key: SENTRY-1767
> URL: https://issues.apache.org/jira/browse/SENTRY-1767
> Project: Sentry
> Issue Type: Sub-task
> Affects Versions: sentry-ha-redesign
> Reporter: kalyan kumar kalvagadda
> Assignee: Na Li
> Fix For: sentry-ha-redesign
>
>
> Suppose that we have two HMSFollower starting up and for some reason they
> both decided they are masters. They both have initial value for the
> notification ID and both try to get full snapshot.
> This should be avoided.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
