Hello Stephen
On 16/1/19 10:27 μ.μ., Stephen Frost wrote:
> Greetings,
>
> * Achilleas Mantzios (achill@matrix.gatewaynet.com) wrote:
>> I had this error today with postgresql 10.6 and pgbackrest 2.07. I was running test backups and during a backup
>> The command given to the repo host (pg1-host=10.9.0.77) was :
>> pgbackrest@smadb2cs:~$ pgbackrest --stanza=dynacomtest backup
>> I noticed that the (two) ssh login sessions to the repo host just froze. No disconnections, just froze. So some
networkproblem surely occurred.
>> On the repo host pgbackrest log I got :
>> 2019-01-16 12:20:38.372 P00 ERROR: [042]: unexpected EOF reading line from remote process on '10.9.0.77'
>> on the pg host (10.9.0.77) I noticed that the pgbackrest backup processes
>> had died, the only ones running was spawn from the archive-push , however on
>> the repo host the pgbackrest processes were alive, when in fact the backup
>> was stalled.
>> On the repo I got :
>> root@smadb2cs:/home/pgbackrest# netstat -apn | grep 10.9.0.77
>> tcp 0 88 10.9.1.22:43837 10.9.0.77:22 ESTABLISHED 11918/ssh
>> tcp 0 88 10.9.1.22:43838 10.9.0.77:22 ESTABLISHED 11920/ssh
>> tcp 0 88 10.9.1.22:43839 10.9.0.77:22 ESTABLISHED 11919/ssh
>> tcp 0 88 10.9.1.22:43840 10.9.0.77:22 ESTABLISHED 11922/ssh
>> root@smadb2cs:/home/pgbackrest# netstat -anp --timers | grep 0.77
>> tcp 0 88 10.9.1.22:43837 10.9.0.77:22 ESTABLISHED 11918/ssh on (65.50/12/0)
>> tcp 0 88 10.9.1.22:43838 10.9.0.77:22 ESTABLISHED 11920/ssh on (69.60/12/0)
>> tcp 0 88 10.9.1.22:43839 10.9.0.77:22 ESTABLISHED 11919/ssh on (65.50/12/0)
>> tcp 0 88 10.9.1.22:43840 10.9.0.77:22 ESTABLISHED 11922/ssh on (61.40/12/0)
> This looks like the kernel thinks the connections are still alive.
>
>> In order to start a new backup I had to kill the 4 processes.
>>
>> Is this behavior normal ? Would the 4 processes be killed due to KEEPALIVE settings eventually ?
> Yes, they almost certainly would have timed out eventually- but how
> long were you already waiting..? The timeouts in pgbackrest are not
> terribly long, in general.
Looking at the log I waited 9 minutes, before manually killing.
>
> Thanks!
>
> Stephen