HN Explorer

Ask HN: why are requests pushed to heroku dynos?

3 points by radiospiel ↗ HN
The last days saw numerous discussions on HN on intelligent vs random routing on heroku. One thing that I don't understand is why requests are routed to dynos when they are entering the heroku mesh, and in the case of random routing might get stuck behind a slow request.

What I don't: why are requests pushed to dynos, and why doesn't a dyno actively pulls a request from some kind of shared queue after it is done processing the previous request.

Can someone explain this?

1 comment

[ 3.2 ms ] story [ 10.8 ms ] thread
iSloth
I'm fairly sure that moving to a pull queue would generate more overhead for each request, granted it would probably solve their current issues. However IMO sticking with a push setup and solving the brokenness would be a higher performance solution, mainly due to the simplicity.

It really sounds like they need to shard their current setup as a minimum, and preferably find some way to synchronise the session table over the routes within a mesh, so any one router is aware which dyno's are blocking/available for that queued traffic.