Ardent Performance Computing

Latest Post

PostgresConf 2019 Training Days

It feels like PostgresConf in New York is in full swing, even though the main tracks haven’t even started yet!

(Oh, and by the way, as of this morning I heard there are still day-passes available for those who haven’t yet registered for the conference… and then you can come hear a great session about Wait Events in PostgreSQL this Thursday at 4:20pm!)

The first two days of PostgresConf are summits, tutorials and training sessions. A good chunk of my day today was helping out with Scott Mead’s intensive 3 hour hands-on lab Setting up PostgreSQL for Production in AWS – but outside of that I’ve managed to drop in to a number of other sessions that sounded interesting. I did my best to take down some notes so I could share a few highlights.

Monday March 18

Personally, my favorite session on Monday was Brent Bigonger’s session.  He’s a database engineer at Amazon who was involved in migrating their Inventory Management System to Aurora PostgreSQL. I always love hearing good stories (part of why I’ve always been a fan of user groups) – this presentation gave a nice high level overview of the business, a review of the planning and execution process for the migration, and lots of practical lessons learned.

• Some of the tips were things people are generally familiar with – like NULLs behaving differently and the importance of performance management with a tool like Performance Insights.
• My favorite tip is getting better telemetry by instrumenting SQL with comments (SELECT /* my-service-call-1234 */ …) which reminded me of something I also read in Baron Sc​hwartz’s recently updated e-book on observable systems: “including implicit data in SQL.”
• A great new tip (to me) was the idea of creating a heartbeat table as one more safety check in a replication process.  You can get a sense for lag by querying the table and you can also use it during a cutover to get an extra degree of assurance that no data was missed.
• Another general point I really resonated with: Brent gave a nice reminder that a simple solution which meets the business requirements is better than a sophisticated or complex solution that goes beyond what the business really needs.  I feel tempted on occasion to leverage architectures because they are interesting – and I always appreciate hearing this reiterated!

On the AWS track, aside from Brent’s session, I caught a few others: Jim Mlodgenski giving a deep dive on Aurora PostgreSQL architecture and Jim Finnerty giving a great talk on Aurora PostgreSQL performance tuning and query plan management.  It’s funny, but I think my favorite slide from Finnerty’s talk was actually one of the simplest and most basic; he had a slide that just had high-level list of steps for performance tuning.  I don’t remember the exact list on that slide at the moment, but the essential process: (1) identify to top SQL (2) EXPLAIN to get the plan (3) make improvements to the SQL and (4) test and verify whether the improvements actually had the intended effect.

Other sessions I dropped into:

• Alvaro Hernandez giving an Oracle to PostgreSQL Migration Tutorial.  I love live demos (goes along with loving hands on labs) and so this session was a hit with me – I wasn’t able to catch the whole thing but I did catch a walk-through of ora2pg.
• Avinash Vallarapu giving an Introduction to PostgreSQL for Oracle and MySQL DBAs. When I slipped in, he was just wrapping up a section on hot physical backups in PostgreSQL with the pg_basebackup utility.  After that, Avi launched into a section on MVCC in PostgreSQL – digging into transaction IDs and vacuum, illustrated with block dumps and the pageinspect extension.  The part of this session I found most interesting was actually a few of the participant discussions – I heard lively discussions about what extensions are and about comparisons with RMAN and older versions of Oracle.

Tuesday March 19

As I said before, a good chunk of my morning was in Scott’s hands-on lab. If you ever do a hands-on lab with Scott then you’d better look out… he did something clever there: somewhere toward the beginning, if you followed the instructions correctly, then you would be unable to connect to your database!  Turns out this was on purpose (and the instructions actually tell you this) – since people often have this particular problem connecting when they first start on out RDS, Scott figured he’d just teach everyone how to fix it.  I won’t tell you what the problem actually is though – you’ll have to sign up for a lab sometime and learn for yourself.  :)

As always, we had a lot of really interesting discussions with participants in the hands-on lab.  We talked about the DBA role and the shared responsibility model, about new tools used to administer RDS databases in lieu of shell access (like Performance Insights and Enhanced Monitoring), and about how RDS helps implement industry best practices like standardization and automation. On a more technical level, people were interested to learn about the “pgbench” tool provided with postgresql.

In addition to the lab, I also managed to catch part of Simon Riggs’ session Essential PostgreSQL11 Database Administration – in particular, the part about PostgreSQL 11 new features.  One interesting new thing I learned was about some work done specifically around the performance of indexes on monotonically increasing keys.

Interesting Conversations

Of course I learned just as much outside of the sessions as I learned in the sessions.  I ended up eating lunch with Alexander Kukushkin who helped facilitate a 3 hour hands-on session today about Understanding and implementing PostgreSQL High Availability with Patroni and enjoyed hearing a bit more about PostgreSQL at Zalando. Talked with a few people from a government organization who were a long-time PostgreSQL shop and interested to hear more about Aurora PostgreSQL. Talked with a guy from a large financial and media company about flashback query, bloat and vacuum, pg_repack, parallel query and partitioning in PostgreSQL.

And of course lots of discussions about the professional community. Met PostgresConf conference volunteers from California to South Africa and talked about how they got involved in the community.  Saw Lloyd and chatted about the Seattle PostgreSQL User Group.

The training and summit days are wrapping up and now it’s time to get ready for the next three days: keynotes, breakout sessions, exposition, a career fair and more!  I can’t wait.  :)