postgres does not work with multiple connections

[gordon chung]

postgres throws SSL errors when using multiple connections.

2014-03-26 14:51:00.659 | *** Not Whitelisted *** 2014-03-26 14:15:51.538 27982 ERROR ceilometer.dispatcher.database [req-2153b896-eb17-478a-90e4-8e168c806542 admin - - - -] Failed to record metering data: (OperationalError) SSL SYSCALL error: EOF detected

http://logs.openstack.org/16/81216/2/check/check-tempest-dsvm-postgres-full/988de01/console.html#_2014-03-26_14_51_00_070

UPDATE::
this appears to work with multiple individual collectors (ie. define two collector services with one worker each)... it appears the way ProcessLauncher creates service via eventlet causes issues with postgres but not mysql/mongo.

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/83215
Committed: https://git.openstack.org/cgit/openstack/ceilometer/commit/?id=965539ba40bd2930f440f6384b9a629d5736b5c2
Submitter: Jenkins
Branch: master

commit 965539ba40bd2930f440f6384b9a629d5736b5c2
Author: Gordon Chung <email address hidden>
Date: Wed Mar 26 16:57:15 2014 -0400

    enable a single worker by default

    there are issues with postgresql and multiple connections. change
    default to a single worker to still allow multiple workers for
    backends which support it.

    Partial-Bug: #1298073
    Change-Id: I1bf256f333ad8614bcae2c74dc0544553bffbbe7

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/83140
Committed: https://git.openstack.org/cgit/openstack/ceilometer/commit/?id=d21a2d2e363d85f8e32ec25b6245dd475dedf7e8
Submitter: Jenkins
Branch: master

commit d21a2d2e363d85f8e32ec25b6245dd475dedf7e8
Author: Gordon Chung <email address hidden>
Date: Wed Mar 26 13:42:08 2014 -0400

    SSL errors thrown with Postgres on multi workers

    database connections were made on service initialisation and not
    on process start. this led to all processes of a service sharing
    the same connection. in this patch, we create unique
    dispatch_manager to handle a different engine for each process.

    we keep default to a single worker as schema needs to be updated
    to avoid possible race/deadlock conditions

    Closes-Bug: #1298073
    Change-Id: I2ad05e2085c0c0f78653c6354d301d18b8dee121