Month: November 2021

Milliseconds vs Microseconds

I recently ran into an issue that I didn’t see coming. I was creating records, and then querying for them from the database.

I knew that they were created very close together in time, but they were created in two separate steps, so I knew that the timestamps would be different.

Only when I queried for them and compared the timestamps in my tests, I was getting the same timestamp back on both records.

After some digging I realized that the records had different timestamps in the database…where things were being measured in microseconds, but Javascript only tracks down to the millisecond, so under certain circumstances I was seeing one record round up and the other record round down such that the two of them were showing the exact same timestamp — to the millisecond on my in-memory object in Javascript.

I never would have expected to be in a situation where I needed or cared about more precision than 1/1000th of a second, but there you have it.

Postgres Create & Return Order

I recently wrote some tests that ended up being flakey. After some investigation, I realized that the objects I was getting back from my Postgres database weren’t always in the order that I was expecting them to be.

I was using a single update or insert statement to create several records at the same time and then querying that same table. My initial experiments had indicated that the objects were always being returned in the same order that I passed them into the insert/update statement, but as I grew the number of tests using that assumption and ran them dozens of times, it became apparent that this assumption is flawed.

It appears that Postgres by default generally saves the objects in the order they are passed in the insert/update statement, and that it generally returns them in that same order, but sometimes it doesn’t.

I haven’t pinned down whether the breakdown is in saving, the retrieving, or both. My money is that it’s on the saving side of things, but it doesn’t actually matter for my purposes. For now, I’ve been updating these tests to return in a given order, ordering by something other than the created_at timestamp.

A good learning for me there about not leaving anything to chance when it comes to a test. Testing…one of the few places where being more explicit is better.