[pgpool-general: 6573] Re: Pgpool query

Lakshmi Raghavendra lakshmiym108 at gmail.com
Mon May 27 19:45:25 JST 2019


Hi,

   i have configured a stream replication pgpool and using watchdog for
failovers.
I often see the below messages in the log:

2019-05-27T10:34:51+00:00  process[6714]: [15-1] 2019-05-27 10:34:51: pid
6714: LOG:  Replication of node:1 is behind 671088848 bytes from the
primary server (node:0)
2019-05-27T10:34:51+00:00  process[6714]: [15-2] 2019-05-27 10:34:51: pid
6714: CONTEXT:  while checking replication time lag
2019-05-27T10:34:51+00:00  process[6714]: [16-1] 2019-05-27 10:34:51: pid
6714: LOG:  Replication of node:2 is behind 1124073680 bytes from the
primary server (node:0)
2019-05-27T10:34:51+00:00  process[6714]: [16-2] 2019-05-27 10:34:51: pid
6714: CONTEXT:  while checking replication time lag
2019-05-27T10:35:01+00:00  process[6714]: [17-1] 2019-05-27 10:35:01: pid
6714: LOG:  Replication of node:1 is behind 671088848 bytes from the
primary server (node:0)
2019-05-27T10:35:01+00:00  process[6714]: [17-2] 2019-05-27 10:35:01: pid
6714: CONTEXT:  while checking replication time lag
2019-05-27T10:35:01+00:00  process[6714]: [18-1] 2019-05-27 10:35:01: pid
6714: LOG:  Replication of node:2 is behind 1124073680 bytes from the
primary server (node:0)
2019-05-27T10:35:01+00:00  process[6714]: [18-2] 2019-05-27 10:35:01: pid
6714: CONTEXT:  while checking replication time lag

Questions :

1. Does the above statement mean when failover happens and node  1 / 2 is
choosen as the primary postgres master, since there is some replication
lag, some data would be lost?

2. Please let me know how to over come this ?

Thanks And Regards,

   Lakshmi Y M


On Fri, May 24, 2019 at 12:18 PM Tatsuo Ishii <ishii at sraoss.co.jp> wrote:

> >> Ok. So the reason behind the ask was if pgpool is not going to retry the
> > promote on other nodes, I was thinking i can just trigger a force promote
> > there by failover and follow master will be run by pgpool itself and i
> dont
> > have to do a failover by myself and run pcp_recovery on all the nodes
>
> Probably you misunderstand what Pgpool-II is expected to do. Pgpool-II
> does nothing with PostgreSQL promotion. It's completely the script
> writer's responsibility to handle the promotion retrying if
> neccessary.
>
> > Regarding 5. It's normal. Don't worry.
> >
> > When a standby PostgreSQL node is added, existing connections to
> > Pgpool-II are kept (thus do not use the new standby for load
> > balacing). After the session ends, Pgpool-II child process checks
> > whether the failback has happend. If happend, exits itself so that new
> > process is spwan to reflect the fact that new standby node has been
> > added (thus it can use the new standby for load balancing).
> >
> >
> >
> >
> >> Thanks. Per my understanding, So restart happens only when a new standby
> > is added is it. Because i see this frequent restarts happening even when
> > there is no fallback.
>
> The failbacke event detection could happen long after the failback
> happend if a user keeps on connection to Pgpool-II.
>
> > Also could you please share some light on when the failback_command gets
> > executed? I have never seen the command being called.
>
> It is called when pcp_attach_node or pcp_recovery_node get
> executed. Also it could happen when follow_master_command gets
> executed if it executes pcp_recovery_node inside the command.
>
> >
> > Thanks And Regards,
> >
> >    Lakshmi Y M
> >
> > On Tue, May 21, 2019 at 8:02 PM Lakshmi Raghavendra <
> lakshmiym108 at gmail.com>
> > wrote:
> >
> >> Hi,
> >>
> >>    I had couple of queries as below :
> >>
> >> On Thu, May 16, 2019 at 2:17 AM Muhammad Usama <m.usama at gmail.com>
> wrote:
> >>
> >>>
> >>>
> >>> On Fri, May 10, 2019 at 6:38 PM Lakshmi Raghavendra <
> >>> lakshmiym108 at gmail.com> wrote:
> >>>
> >>>> Hi,
> >>>> I am trying pgpool for automatic faliover on my postgresql cluster
> using
> >>>> the watchdog feature
> >>>> Wanted to know a couple of things
> >>>> 1. Are there any hooks when pgpool re-elects next pgpool master ?
> wanted
> >>>> to run some customization during this time
> >>>>
> >>>
> >>> Pgpool executes the user provided commands at the time of acquiring and
> >>> releasing of virtual IP by master pgpool node. You can
> >>> configure wd_escalation_command and wd_de_escalation_command
> configuration
> >>> parameters to provide the custom command or script.
> >>> Pgpool which executed wd_escalation_command command when it gets
> elected
> >>> as a master and performs the escalation
> >>> and similarly when the it node will resign as a master the
> wd_de_escalation_command
> >>> command gets executed,
> >>>
> >>
> >> 1. I tried using the  above wd_escalation_command and
> wd_de_escalation_command.
> >> I have a 3 node cluster, Observed that the escalation command will be
> >> triggered if there are at least 2 nodes alive in the pgpool cluster.
> >> If there is only master pgpool alive with no slave nodes, the command is
> >> never initiated. Is this behavior expected ?
> >>
> >> 2. Wanted to understand the significance of VIP, Is there any issue
> caused
> >> if I dont use the VIP in a 3-node pgpool cluster.
> >>
> >> Please let me know.
> >>
> >> Thanks And Regards,
> >>
> >>    Lakshmi Y M
> >>
> >> On Thu, May 16, 2019 at 2:17 AM Muhammad Usama <m.usama at gmail.com>
> wrote:
> >>
> >>>
> >>>
> >>> On Fri, May 10, 2019 at 6:38 PM Lakshmi Raghavendra <
> >>> lakshmiym108 at gmail.com> wrote:
> >>>
> >>>> Hi,
> >>>> I am trying pgpool for automatic faliover on my postgresql cluster
> using
> >>>> the watchdog feature
> >>>> Wanted to know a couple of things
> >>>> 1. Are there any hooks when pgpool re-elects next pgpool master ?
> wanted
> >>>> to run some customization during this time
> >>>>
> >>>
> >>> Pgpool executes the user provided commands at the time of acquiring and
> >>> releasing of virtual IP by master pgpool node. You can
> >>> configure wd_escalation_command and wd_de_escalation_command
> configuration
> >>> parameters to provide the custom command or script.
> >>> Pgpool which executed wd_escalation_command command when it gets
> elected
> >>> as a master and performs the escalation
> >>> and similarly when the it node will resign as a master the
> wd_de_escalation_command
> >>> command gets executed,
> >>>
> >>>
> >>>
> http://www.pgpool.net/docs/latest/en/html/runtime-watchdog-config.html#CONFIG-WATCHDOG-ESCALATION-DE-ESCALATION
> >>>
> >>>
> >>>> 2. Will the VIP get assigned ony if there are more than 1 node present
> >>>> in the pgpool cluster? I had 3 nodes where i had pgpool running. When
> the
> >>>> 1st and 2nd node's pgpool was shut i was expecting the 3rd node to
> acquire
> >>>> the VIP but it didn't happen. And if my understanding was right i was
> >>>> thinking of using the VIP in my database connection string (Since it
> will
> >>>> always be with pgpool master who can connect to my postgresql
> primary). Now
> >>>> if the 3rd node is not acquiring the VIP, i could not use it my
> connection
> >>>> string. Correct me if my understanding is wrong
> >>>>
> >>>
> >>> master pgpool only acquires the VIP when the quorum exists ( minimum
> 50%
> >>> of nodes are reachable). This is done by Pgpool to guard against the
> >>> split-brain syndrome, which could happen
> >>> otherwise in case of network partitioning. So if you have 3 Pgpool
> nodes
> >>> configured than the VIP will only get assigned on master node when
> at-least
> >>> 2 Pgpool nodes are alive and reachable.
> >>> but in case of 2 node configuration, only 1 node is required to ensure
> >>> the quorum and in that case even if the single node is alive, it will
> get
> >>> the VIP
> >>>
> >>> Thanks
> >>> Best Regards
> >>> Muhammad Usama
> >>>
> >>>>
> >>>>
> >>>> Thanks in advance
> >>>>
> >>>>    Lakshmi Y M
> >>>> _______________________________________________
> >>>> pgpool-general mailing list
> >>>> pgpool-general at pgpool.net
> >>>> http://www.pgpool.net/mailman/listinfo/pgpool-general
> >>>>
> >>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.sraoss.jp/pipermail/pgpool-general/attachments/20190527/feb822e8/attachment.html>


More information about the pgpool-general mailing list