Since upgrading from 0.27.x to 0.28.x, we are seeing extremely long load times for the initial page load of the Metabase home screen (where you land when you are already logged in). Currently over 1 min to load the home screen right now on the first attempt. Subsequent requests by the same user load quickly.
We tried bumping the database instance size, no effect.
Running Metabase on AWS ElasticBeanstalk.
Any ideas?
Thanks for reporting this.
What region are you in?
We’ve seen this come and go on one of our own instances but can’t seem to unravel exactly where the delay is coming from. Do you have access to the beanstalk logs?
Yes. We are in us-east-1. Do you just want a full dump of the beanstalk logs, or ??
Can you screenshot the “Network” panel of the developer tools when you load the homepage up? A starting point would be seeing which file load the delay is coming from.
Attaching Network screenshots (had to break into two that overlap a bit). Also blacked-out our true URL for security reasons.
All of the files seem to load in 1-2s once they get going.
Are you using route 53 for your domain name?
Yes using Route53.
Note – this didn’t start happening until we upgraded to v.0.28.x. Everything was fine on 0.27.x. No change to our infrastructure at all.
Practically speaking, not a whole lot changed between 0.27 and 0.28 in terms of frontend bundle size or anything related to how that file is served up.
In our case, I was’t able to repro this issue on a local server, heroku, or a raw docker image on ec2. There’s something beanstalk specific going on =(
One thing to note that under the hood there are multiple versions of beanstalk floating around, and an upgrade can result in the environment using a different version. That’s my best guess as to why this started to affect you now.
One annoying but fairly sure fire way to fix this is to snapshot the RDS instance, spin up a new one and migrate off of beanstalk to a raw docker image. You lose some of the niceties of beanstalk, such as autorestart, etc, but it does tend to have less magic going on.
We’re trying to figure out whether we want to keep recommending beanstalk…
Interesting. Ok maybe we will try just docker then. If we spin up another brand new beanstalk env with the 0.27 version, it works fine again, just FYI.
