Re: Libpq support to connect to standby server as priority

Laurenz Albe <laurenz.albe@cybertec.at>

From: Laurenz Albe <laurenz.albe@cybertec.at>
To: Haribabu Kommi <kommi.haribabu@gmail.com>
Cc: Jing Wang <jingwangian@gmail.com>, Pg Hackers <pgsql-hackers@postgresql.org>, "Tsunakawa, Takayuki" <tsunakawa.takay@jp.fujitsu.com>
Date: 2018-07-16T14:42:23Z
Lists: pgsql-hackers
Haribabu Kommi wrote:
> > On Wed, Jul 4, 2018 at 11:14 PM Laurenz Albe <laurenz.albe@cybertec.at> wrote:
> > > Haribabu Kommi wrote:
> > >  
> > > - I think the construction with "read_write_host_index" makes the code even more
> > >   complicated than it already is.
> > > 
> > >   What about keeping the first successful connection open and storing it in a
> > >   variable if we are in "prefer-read" mode.
> > >   If we get the read-only connection we desire, close that cached connection,
> > >   otherwise use it.
> > 
> > Even if we add a variable to cache the connection, I don't think the logic of checking
> > the next host for the read-only host logic may not change, but the extra connection
> > request to the read-write host again will be removed.
> 
> I evaluated your suggestion of caching the connection and reuse it when there is no
> read only server doesn't find, but I am thinking that it will add more complexity and also
> the connection to the other servers delays, the cached connection may be closed by
> the server also because of timeout.
> 
> I feel the extra time during connection may be fine, if user is preferring the prefer-read
> mode, instead of adding more complexity in handling the cached connection? 
> 
> comments?

I tested the new patch, and it works as expected.

I don't think that time-out of the cached session is a valid concern, because that
would have to be a really short timeout.
On the other hand, establishing the connection twice (first to check if it is read-only,
then again because no read-only connection is found) can be quite costly.

But that is a matter of debate, as is the readability of the code.

Since I don't think I can contribute more to this patch, I'll mark it as
ready for committer.

Yours,
Laurenz Albe


Commits

  1. Make test_target_session_attrs more robust against connection failure.

  2. Extend the abilities of libpq's target_session_attrs parameter.

  3. Mark default_transaction_read_only as GUC_REPORT.

  4. Introduce a new GUC_REPORT setting "in_hot_standby".

  5. Avoid spamming the client with multiple ParameterStatus messages.

  6. Avoid harmless Valgrind no-buffer-pin errors.

  7. Restructure libpq code to remove some duplicity

  8. libpq: Add target_session_attrs parameter.