Mostly aws media services stuff such as mediaconnect, medialive, some directly using

aws

provider, a couple bits using

aws-native

(which is buggy as heck with media services resources, lots of 500 errors from the aws api) and some using cloudformation.Stack. I'd say about 9 mediapackage channels, 9 medialive channels, 30 or so mediaconnect flows, as well as 2 cloudfront distributions, some roles, some secrets. With all the bits & pieces that wire up the major parts it comes to 250 resources, i'm sure 1/5 to 1/10 of those are my custom components. Most interesting is that the same thing takes 1.5 minutes on my coworkers machine, who is in australia. Our backend is an s3 bucket in australia. I timed a call to fetch the state json file and it takes 6.5 seconds to fetch it on my machine i'm not ready to make a strong statement on what i think i'm observing, take the following with a grain of salt because I don't trust my memory. I have noticed that, depending on what i'm trying to update, the preview portion of

pulumi up

can take anywhere from a minute to 5, then on

yes

the preparatory phase up to 15min where nothing is updated, then finally it begins updating resources and that part appears to go reasonably well although ive noticed that resources may have updated and then pulumi can sit another 20-60 seconds extra. I did notice that when I was fiddling with the cloudformation.Stacks the times swung quite a lot, and somehow I think when it had nothing to update it got slower

i'm not sure if it's a proximity thing or my machine

versions are:

pulumi v3.92.0

@pulumi/pulumi: 3.91.1
@pulumi/aws: 6.6.1
@pulumi/aws-native: 0.82.0
@pulumi/awsx: 2.0.3
@pulumi/random: 4.14.0

the terraform provider for media services is very old & incapable, i'm doing a refresher course on golang so i can maybe submit some fixes

and aws's own cloud control api throws 500

so it's gotta be cloudformation

either way it's not meant to take 20+ minutes so something is definitely going wrong, im going to do a trace this afternoon to see if it's doing that goofy thing where it writes to the state file for each resource in series, downloading & uploading it each time

was a bug in previes versions of pulumi, wondering if it's regressed somehow

I guess we don't have any sort of 'active' state system that can exchange diffs instead of the whole state file hey? nvm i'll go read

oh lord 1.5 minutes vs 7.5 for a different stack

zzzz

I'm guessing that we cant manage locks & state in separate backends can we