Just got paged at 2 AM because a batch job was stuck... again. While I was waiting for the logs to stream, I realized I've been letting MeetGeek record our daily standups for months. The summaries are decent, but I have no idea if I'm getting them in 2 minutes or 20.
In our world, latency is a feature. If the summary takes longer than the coffee break after the call, it's useless. My team won't wait.
So, has anyone actually done a proper benchmark? I'm talking:
* **Cold start vs. warm:** First call of the day vs. consecutive calls.
* **Call length:** Does a 15-minute sync process as fast as a 60-minute marathon per minute?
* **API peek:** Any HTTP `Retry-After` headers or queueing behavior in the responses?
My anecdotal experience is all over the place. Sometimes it's there before I've finished typing my follow-up Slack message. Other times, I've already written my own notes and moved three Jira tickets before it pops up.
If no one has numbers, I'm tempted to spin up a test:
* Record a series of dummy calls with `ffmpeg`
* Script the upload via their API
* Measure the delta between `upload completed` and `summary available` webhook
But if someone's already done the grunt work, you'd save me a night shift project. My `terraform apply` queue is long enough.
Pager duty survivor.
NightOps