On Wed, 06 Oct 2010 08:43:44 -0000, Martin Pool <email address hidden> wrote:
> notes from talking this over with spm:
>
> * at the moment you can see in the process listing which branch is
> being accessed; it would be very nice to keep that; but at least we
> should log the pid/user/branch
This isn't affected by this branch, although the process name ends up
being a bit unwieldy:
Maybe inserting a call to setproctitle in become_child would be useful,
especially as we have the dependency already -- one piece of information
we've lost is the user who has connected, although we can probably
figure that out from logs still.
> * can run this on staging, then edge, then lpnet
> * propose changes into lp:lp-production-configs to gradually turn it on
> * might want an rt describing how to monitor the new daemon process,
> which just needs to mention the socket location and the hello command;
> they can write the script from there
> * francis should formally tell tom and charlie we'll be making this change
I still vaguely wonder about controlling this with a feature flag. The
thing is, the codehosting systems don't connect to the database
currently, and that's actually kind of nice in some ways. We could add
an XML-RPC method to query a flag, and have the ssh daemon check that
every 30s or so -- I don't think we want an XML-RPC call delaying the
launch of the lp-servce process on every connection.
On Wed, 06 Oct 2010 08:43:44 -0000, Martin Pool <email address hidden> wrote:
> notes from talking this over with spm:
>
> * at the moment you can see in the process listing which branch is
> being accessed; it would be very nice to keep that; but at least we
> should log the pid/user/branch
This isn't affected by this branch, although the process name ends up
being a bit unwieldy:
21266 pts/1 S+ 0:00 /usr/bin/python /home/mwh/ canonical/ checkouts/ trunk/eggs/ bzr-2.2. 0-py2.6- linux-x86_ 64.egg/ EGG-INFO/ scripts/ bzr launchpad- forking- service --path /var/tmp/ launchpad_ forking_ service. sock BRANCH: ~mark/upstart/ branch
Maybe inserting a call to setproctitle in become_child would be useful,
especially as we have the dependency already -- one piece of information
we've lost is the user who has connected, although we can probably
figure that out from logs still.
> * can run this on staging, then edge, then lpnet
> * propose changes into lp:lp-production-configs to gradually turn it on
> * might want an rt describing how to monitor the new daemon process,
> which just needs to mention the socket location and the hello command;
> they can write the script from there
> * francis should formally tell tom and charlie we'll be making this change
I still vaguely wonder about controlling this with a feature flag. The
thing is, the codehosting systems don't connect to the database
currently, and that's actually kind of nice in some ways. We could add
an XML-RPC method to query a flag, and have the ssh daemon check that
every 30s or so -- I don't think we want an XML-RPC call delaying the
launch of the lp-servce process on every connection.
Cheers,
mwh