a year ago
My DB Queries just turned slow like crazy …
I'm using a DJANGO App (which means I don't have a complete control over the SQL queries executed).
Most of the queries became between 30-150x slower (yes I'm using private URL of the postgresql DB).
From 0.14ms per query to 19-24ms per query.
In total the profiler reports 3534ms spent waiting for the DB against 34ms before … A factor 100x … How can this is be acceptable ??
How can you guys force such a massive change over christmas ? Do you realize how disruptive and time consuming this is to debug while everyone is supposedly off ? I'm very disappointed in Railway, really like the platform, extremely disappointed by the execution. I don't even know if I can fix anything… Railway has become prohibitively slow for Django as a whole.
14 Replies
Photos linked before & after. Number on the right is the time in ms per query
a year ago
The reason this happens is because Railway switched hobby workloads over to metal, but metal doesn't support persistent storage yet so databases remain on GCP, this zone difference causes ~20ms of delay
The delay is temporary though and will be solved once Railway has persistent storage for metal (which I believe is around the corner)
Note, if 20ms of delay is mission critical for your application and hurts your business then it's likely you're not supposed to be on hobby plan in the first place, hobby plan is only meant for hobby-specific workloads. For business workloads you should really be on pro plan
How is it possible that you guys force such a MASSIVE change with no warning and over a time where everybody is off ?
I have seen some terrible engineering and customer service in my life … This is probably in the TOP 5 …
How is the Pro Plan going to fix anything. One of the Railway Employee mentioned being on the Pro is not a long term fix because it will be moved to METAL soon
And 20ms is obviously not mission critical … But 20ms per query is huge ! It translates into 4-5sec / page
a year ago
Railway wouldn't move pro workloads to metal if metal doesn't have support for volumes
Pro plan will fix the current delay because you can then move your workload back to the same zone your database is in
a year ago
little hack: you could probably add a volume to your app so they stay in GCP (metal doesnt have support for volumes), thats not a permanent solution and can break at any time but will give you sometime
@ThallesComH unfortunately looks like it's not possible anymore to create a storage volume
a year ago
You just need to add the volume on the same step as service creation
a year ago
happy to help
a year ago
!s
Status changed to Solved dev • about 1 year ago