public inbox for [email protected]  

help / color / mirror / Atom feed

From: Drouvot, Bertrand <[email protected]>
To: Euler Taveira <[email protected]>
To: PostgreSQL Hackers <[email protected]>
Subject: Re: Move walreceiver state assignment (to WALRCV_STREAMING) in WalReceiverMain()
Date: Mon, 18 Dec 2023 11:18:10 +0100
Message-ID: <[email protected]> (raw)
In-Reply-To: <[email protected]>
References: <[email protected]>
	<[email protected]>

Hi,

On 12/12/23 8:54 PM, Euler Taveira wrote:
> On Tue, Dec 12, 2023, at 12:58 PM, Drouvot, Bertrand wrote:
>> Currently walrcv->walRcvState is set to WALRCV_STREAMING at the
>> beginning of WalReceiverMain().
>>
>> But it seems that after this assignment things could be wrong before the
>> walreicever actually starts streaming (like not being able to connect
>> to the primary).
>>
>> It looks to me that WALRCV_STREAMING should be set once walrcv_startstreaming()
>> returns true: this is the proposal of this patch.
> 
> Per the state name (streaming), it seems it should be set later as you
> proposed, 

Thanks for looking at it!

> however, I'm concerned about the previous state (WALRCV_STARTING). If
> I'm reading the code correctly, WALRCV_STARTING is assigned at
> RequestXLogStreaming():
> 
>                          SetInstallXLogFileSegmentActive();
>                          RequestXLogStreaming(tli, ptr, PrimaryConnInfo,
>                                               PrimarySlotName,
>                                               wal_receiver_create_temp_slot);
>                          flushedUpto = 0;
>                      }
> 
>                      /*
>                       * Check if WAL receiver is active or wait to start up.
>                       */
>                      if (!WalRcvStreaming())
>                      {
>                          lastSourceFailed = true;
>                          break;
>                      }
> 
> After a few lines the function WalRcvStreaming() has:
> 
>      SpinLockRelease(&walrcv->mutex);
> 
>      /*
>       * If it has taken too long for walreceiver to start up, give up. Setting
>       * the state to STOPPED ensures that if walreceiver later does start up
>       * after all, it will see that it's not supposed to be running and die
>       * without doing anything.
>       */
>      if (state == WALRCV_STARTING)
>      {
>          pg_time_t   now = (pg_time_t) time(NULL);
> 
>          if ((now - startTime) > WALRCV_STARTUP_TIMEOUT)
>          {
>              bool        stopped = false;
> 
>              SpinLockAcquire(&walrcv->mutex);
>              if (walrcv->walRcvState == WALRCV_STARTING)
>              {
>                  state = walrcv->walRcvState = WALRCV_STOPPED;
>                  stopped = true;
>              }
>              SpinLockRelease(&walrcv->mutex);
> 
>              if (stopped)
>                  ConditionVariableBroadcast(&walrcv->walRcvStoppedCV);
>          }
>      }
> 
> Couldn't it give up before starting if you apply your patch? My main concern is
> due to a slow system, the walrcv_connect() took to long in WalReceiverMain()
> and the code above kills the walreceiver while in the process to start it.

Yeah, so it looks to me that the sequence of events is:

1) The startup process sets walrcv->walRcvState = WALRCV_STARTING (in RequestXLogStreaming())
2) The startup process sets the walrcv->startTime (in RequestXLogStreaming())
3) The startup process asks then the postmaster to starts the walreceiver
4) Then The startup process checks if WalRcvStreaming() is true

Note that 3) is not waiting for the walreceiver to actually start: it "just" sets
a flag and kill (SIGUSR1) the postmaster (in SendPostmasterSignal()).

It means that if the time between 1 and 4 is <= WALRCV_STARTUP_TIMEOUT (10 seconds)
then WalRcvStreaming() returns true (even if the walreceiver is not streaming yet).

So it looks to me that even if the walreceiver does take time to start streaming,
as long as the time between 1 and 4 is <= 10 seconds we are fine.

And I think it's fine because WalRcvStreaming() does not actually "only" check that the
walreceiver is streaming but as its comment states:

"
/*
  * Is walreceiver running and streaming (or at least attempting to connect,
  * or starting up)?
  */
"

Regards,

-- 
Bertrand Drouvot
PostgreSQL Contributors Team
RDS Open Source Databases
Amazon Web Services: https://aws.amazon.com

view thread (3+ messages)

reply

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Reply to all the recipients using the --to and --cc options:
  reply via email

  To: [email protected]
  Cc: [email protected], [email protected], [email protected]
  Subject: Re: Move walreceiver state assignment (to WALRCV_STREAMING) in WalReceiverMain()
  In-Reply-To: <[email protected]>

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox