Notes from the presentation tonight

Here's the notes I took; they're mostly a stream-of-consciousness "that
seems important" but I tried to make a coherent list of takesaway at
the bottom...

Provisioned iOPS (better I/O speed) is only for larger databases;
likewise encryption

DB Instance class codes: t = free eligible m = memory-optimized or maybe
median cr = beastly, may stand for cray, but that might have been a joke
r = mem (definitely) and io (maybe) optimized

You have no pg_hba, which means none of the plugins meant to work with
it (but you have AWS security settings).

Backup and Maintenance windows mustn't overlap

postgresql.conf is sorta available in the form of the "parameter group."
Values from postgresql.conf may not translate directly; e.g. 0 instead
of "off"; no specifying units for some values. Keep the postgresql.conf
docs handy

Cloudwatch metrics are more like system metrics than database metrics.
Database metrics are harder but Stackdriver provides them.

Logs go Where AWS Says They Go. PGBadger needs extra work to work with
AWS's log-line format

Prefer the AWS cli over the specific RDS cli

point-in-time recovery is really fast and lovely

reliable and easy failover protection

AWS forums are pretty well-monitored by AWS reps; may even be faster
than a support call

RDS seems cheaper at first compared to your own hardware, but seems to
become more expensive at about a year. But maybe you have to buy new
hardware after two years, and the comparison doesn't include the cost of
electricy or bandwidth. And AWS has legit clustering, load balancing,
and and and...

Reliable/easy failover and new database creation 🌟💯😂

"There's this stuff that happens that doesn't happen when I have my
own server."

Extension availability: You might think you can `select * from
pg_available_extensions order by name` but no instead, `SHOW
rds.extensions`

Takesaway:
* This is not postgres, it's postgres-ish
* You are not the database superuser
* This is not your system
* You have to view these as ephemeral
> * They will failover, fail, or disappear completely, and you don't
> get to know why
> * Your backups are also ephemeral; autosnapshots are destroyed when
> you destroy an instance
> * If you want your backups in a separate AZ, you have to make that
> happen yourself
* Patches are applied for you; OS upgrades don't cause downtime; PG
upgrades *do*.
* Gabrielle doesn't need an excuse to drink more; she has plenty

--
Andrew