Re: error handling in RegisterBackgroundWorker

On Tue, Apr 11, 2017 at 10:13 PM, Noah Misch <noah(at)leadboat(dot)com> wrote:
> On Tue, Apr 11, 2017 at 11:33:34AM -0400, Tom Lane wrote:
>> Peter Eisentraut <peter(dot)eisentraut(at)2ndquadrant(dot)com> writes:
>> > I think there is no clear agreement here, and no historically consistent
>> > behavior. I'm prepared to let it go and cross it off the list of open
>> > items. I believe we should keep thinking about it, but it's not
>> > something that has to hold up beta.
>>
>> Agreed, this doesn't seem like a must-fix-for-beta consideration.
>
> Definitely not a beta blocker, agreed. Would it be okay to release v10 final
> with the logical replication launcher soft-failing this way?

I'm not really in favor of making a behavior change here, so it would
be fine with me. Obviously people who think the behavior should be
changed are more likely to disagree with that idea. Maybe in the long
run we should have a command-line option (not a GUC) that's like...

postgres --soldier-on-valiently

...and then when that's not supplied we can die but when it is
supplied we persist in spite of all failures that are in any way
recoverable. However, I think that's really a new development effort,
not a cleanup item for v10. I share Tom's concern to the effect that
single-user mode is a really awful way to try to recover from
failures, so we should avoid decisions that force people into that
recovery mode more often.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company