Re: Using PostgreSQL for service discovery and health-check

On Thu, Feb 9, 2023 at 4:46 PM David G. Johnston <david(dot)g(dot)johnston(at)gmail(dot)com>
wrote:

> On Thu, Feb 9, 2023 at 8:30 AM Dominique Devienne <ddevienne(at)gmail(dot)com>
> wrote:
>
>> That's basically my plan. Now come the questions:
>> 1) will updating a row every second (for example) create issues?
>> 2) if yes to #1 above, what would be good mitigation tactics? Use
>> different table for service vs heartbeat? Special kind of table? or
>> configured in a particular way?
>> 3) if a service crashes, it won't remove its row(s), obviously. What kind
>> of mechanism exists to "reap" "zombie" services?
>> 4) Related to #3 above, I think built-in "cron"-like services are only
>> available via extensions, not in PostgreSQL proper. Why? Seems like such an
>> essential service.
>> 5) Which cron-like extension to use? Especially since we run both on-prem
>> but also in managed-PostgreSQL on the cloud?
>>
>>
> You can probably get good mileage from CREATE UNLOGGED TABLE.
>

Thanks. Although I guess the fact it's not replicated to standby servers
could be a problem?
That's not something we test now, and also something the on-prem DBA (i.e.
a client of ours)
might want to setup on his/her own, on top of our client/server arch I
guess.

I have no experience with stand-bys (replication) and HA in PostgreSQL.
Would having the main service table be a regular one, and the
service_heartbeat be an unlogged one be replication friendly?
I.e. if fail over to the stand-by happens, the service table is still there
and populated, but the service_heartbeat is empty, but
then the services would start populating it "transparently" no? I.e. would
using 2 tables instead of 1 be a better design?