SpookyJumpyBeans commented on PR #8898: URL: https://github.com/apache/hbase/pull/8898#issuecomment-7070225298
@charlesconnell, thanks for taking a look! I agree a real clock problem should be visible rather than hidden. My concern with failing startup is that this case doesn't need a clock going backwards on one host. It only needs the backup master's clock to be behind the old master's by more than the time between the two becoming active, which in a fast failover can be a few seconds. HBase already accepts that much skew between servers: `ServerManager.checkClockSkew` only warns above `hbase.master.warningclockskew` (10s) and rejects above `hbase.master.maxclockskew` (30s). If the backup master refused to start in that case, the cluster would be left without an active master during the failover. How about doing both: if the stored active time is ahead of the new master's clock by more than `hbase.master.maxclockskew`, fail startup with an error that says the masters' clocks are out of sync, and below that use the stored time plus one and log a warning, as now. That keeps the behavior consistent with the existing region server check. If you'd rather always fail, I can change it to that instead. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
