Note that this feature in Sentinel would not avoid this problem of persistence-less masters: the slaves will connect back and will empty. In the documentation, what we do is to advise about to configure persistence-less setups so that on restart the machine does not automatically restarts the Redis instance, so that instances cannot recover from reboot. Now that AOF is fixed yes, a master configured to return back without data is dangerous. Reason is that it may have less data but it may very well have more Master is working again, why to trigger a failover at this point? Only Instance the non-availability time configured was longer, now the Reboot when the master is already available, so in the first For this reason toįailover on reboot by default looks a bit too harsh, because we detect Kernel kills Redis for OOM, the Virtual Machine or computer host isĬleanly shutdown and restarted by the provider. Examples: Redis crashes, Redis is manually SHUTDOWN, the Way but sudden kernel crash or power outage, the master should haveĪll the data. However writing just as in write(2) call, Returns, after serving the clients that were already ready to serve, ![]() Is called in the beforeSleep() function, so every time the event loop Ok I said the truth :-) Basically the function flushAppendOnlyFile() [This is the moment when after writing a sentence you go to check the Only in a very odd condition of an immediate power outage basically,īecause while fsync() happens every second, the write to flush data on Things are a bit more complex, copying and pasting what I wrote: We had this exact conversation a few minutes ago in an internal chat about this issue.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |