public inbox for [email protected]
help / color / mirror / Atom feedFrom: George Neuner <[email protected]>
To: [email protected] <[email protected]>
Cc: [email protected]
Subject: Re: Unlogged tables
Date: Wed, 9 Aug 2017 15:52:26 -0400
Message-ID: <[email protected]> (raw)
In-Reply-To: <BN4PR15MB0547F5864AD9691E31458FE1858B0@BN4PR15MB0547.namprd15.prod.outlook.com>
References: <[email protected]>
<BN4PR15MB0547F5864AD9691E31458FE1858B0@BN4PR15MB0547.namprd15.prod.outlook.com>
List-Unsubscribe: <mailto:[email protected]?body=unsub%20pgsql-performance>
Please don't top post.
On 8/9/2017 2:30 PM, [email protected] wrote:
> > On 8/9/2017 2:17 PM, [email protected] wrote:
>
> >> On Wed, 9 Aug 2017 09:14:48 -0700, Jeff Janes <[email protected]> wrote:
>
> >> Why doesn't the Windows scheduled shutdown signal postgres to shutdown
> >> cleanly and wait for it to do so? That is what is supposed to happen.
>
> > Windows *does* signal shutdown (and sleep and hibernate and wakeup).
> > pg_ctl can catch these signals only when running as a service ... it
> > will not catch any system signals when run as an application.
>
> Ok, I am not sure. I run Postgres as a service, and when my Windows
> rebooted after a patch, UNLOGGED tables were cleaned... maybe the
> patch process in Windows messed something up, I don't know.
Hmm. Do you have checkpoint intervals set very long? Or do you have
the Windows shutdown delay(s) set short?
Data in unlogged tables persists only AFTER a checkpoint ... if the
tables had been written to and were "dirty", and the system went down
before the shutdown checkpoint (or before the shutdown checkpoint
completed), then the tables would be truncated at the next startup.
Service control in Windows is very different from Unix/Linux, and
Windows is not completely POSIX compatible. I develop software for
Windows and Linux, but I only use Postgresql. Postgresql was written
originally for Unix and it is possible that the Windows version is not
doing something quite right.
I took a quick glance at the source for pg_ctl: SERVICE_CONTROL_SHUTDOWN
and SERVICE_CONTROL_STOP both just set an shared event to notify the
writer processes to terminate. Offhand I don't see where pg_ctl -
running as a service - is waiting for the writer processes to actually
terminate ( it does wait if run from the command line ). It's possible
that your system shut down too quickly and the WAL writer was killed
instead of terminating cleanly.
Just FYI, re: Postgresql as a user application.
Windows doesn't send *signals* (ala Unix) at all ... it is message
based. The control messages are different for applications and services
- e.g., WM_SHUTDOWN is sent to applications, SERVICE_CONTROL_SHUTDOWN is
sent to services. In order for an application to catch a message, it
must create a window.
pg_ctl is a command line program which does not create any windows (in
any mode). It was designed to enable it to run as a service, but when
run as a user application it will can't receive any system messages.
The user *must* manually stop a running database cluster before shutting
down or sleeping.
George
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]
Subject: Re: Unlogged tables
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