Aurora PostgreSQL 9.6.12: Segmentation fault and restart

Question

Large production database running stable on PostgreSQL 9.6.6/8 for a for a year.  Today upgraded to 9.6.12 and in the first few hours encountered two segmentation faults causing the database to restart.  Generally …

```
LOG: Segmentation fault
DETAIL: The postmaster has commanded this server process to roll back the current transaction and exit, because another server process exited abnormally and possibly corrupted shared memory.
FATAL:  Can't handle storage runtime process crash
LOG:  database system is shut down
FATAL:  the database system is in recovery mode
```

The same query was logged in both case (basic INNER JOINs).  
  
Is this a known issue?  Any advice to bypass it?    
  
Dump/Restore is prohibitive.  I'd like to get ahead of this before the work-week starts.

Answer

That's great news!  Once deployed, I'll recheck our reproducible case and report back quickly.

Answer

The issue appears to be fully resolved.  
  
Thanks for the special support.  Much appreciated.

Answer

Hello AWS,  
  
Check my thread,. We are on Aurora 10.7 and we are running into this all the time, at few times a week. It just bought down our live site again.  
  
https://forums.aws.amazon.com/thread.jspa?threadID=303997

Answer

Hello,  
  
We just upgraded from 9.6.8 to 9.6.12 last night, and we immediately started seeing crashes with a frequency of about once/hour.  
  
Is there another fix/upgrade available for the 9.6-compatible Aurora series?

Answer

We copied the db to a test instance and can reproduce the segfault with a one-line SELECT statement.

Answer

Hi Northrock.  I am a development manager for Aurora PostgreSQL.  We are very sorry for the issue you have encountered on 9.6.12.  Our engineers have identified the problem and we have prepared a patch release.  This is being deployed to our production regions now and will appear as available maintenance for your cluster once it reaches your region.    
  
If you would like to send me a pm with your region and cluster name we can provide some additional options in terms of patching.

Answer

Can we grab a stack trace?

Answer

Updating Aurora 1.5.0 to 1.5.1 resolved our issue.

```
SELECT AURORA_VERSION();
```

Answer

Roll out the AURORA update faster, AWS Support!  
  
This same problem started randomly hitting our 9.6 clusters this week and we just can only watch them restart until things clear up.  
  
Typically we start seeing random queries start taking longer than normal and some take 10 or more minutes. Normally the same query would average under 2 seconds.  
  
I've had to delete and reindex some indices due to errors like: Attempting to read past EOF of relation "base/16402/24527244". blockno=6400 nblocks=3834  
  
Aurora is on version 1.5.0 -- which is something only AWS can update.

Answer

We continue to see the identical segfault/restart pattern every 6 hours or so.  
  
We're rebuilding the large indexes and trying to narrow to a reproducible case.  
  
A similar case, but on a different stack: https://github.com/postgrespro/pg_pathman/issues/193

Aurora PostgreSQL 9.6.12: Segmentation fault and restart

相关内容