On 8 May 2018 at 16:15, Oleksandr Shulgin <oleksandr.shul...@zalando.de> wrote: > Hi Hackers, > > I didn't find the original discussion which led to introduction of the > client-side set of keepalive parameters back in [1]. > > The issue I'm facing is that it doesn't seem to be possible to set these > parameters from the environment variables. The block of option > definitions[2] added for these parameters doesn't specify any corresponding > env var names. > > I wonder if this is an oversight or was there a conscious decision not to > allow it? > > To give a specific example, I have a (legacy) Python script which talks to > the database in two ways: by directly using psycopg2.connect(host=..., ) and > also by running some shell commands with psql's \copy + gzip, mainly for > convenience. > > Now when I needed to tune the keepalives_idle value, I had to include it > everywhere a database connection is made: > - For psycopg2 it's straightforward: you just add an extra keyword parameter > to connect(). > - For psql it's trickier, since the only way to achieve this seems to pack > it together with -d as: -d 'dbname=mydb keepalives_idle=NNN'. > > In case of a binary that would amount to recompiling, I guess. I have no > permission to tweak sysctl directly on the host, unfortunately. > > It would be much more convenient to just set the environment variable when > running the script and let it affect the whole process and its children. > > Would a patch be welcome?
I can't really comment on that part, but: PGOPTIONS="-c tcp_keepalives_count=5 -c tcp_keepalives_interval=60" psql 'host=localhost' should enable server-side keepalives. Unsure how much that helps you if you need client-side keepalives too. -- Craig Ringer http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training & Services
