2 things:
1. Stack Overflow now has dark mode: stackoverflow.blog/2020/03/30/int… - after dropping IE 11 support, we could use CSS3 variables and make it all possible.
2. All Stack Overflow servers are running on .NET Core. We anticipate merging to master tomorrow.
It's launch day y'all.
1. Stack Overflow now has dark mode: stackoverflow.blog/2020/03/30/int… - after dropping IE 11 support, we could use CSS3 variables and make it all possible.
2. All Stack Overflow servers are running on .NET Core. We anticipate merging to master tomorrow.
It's launch day y'all.
We found a few minor glitches and one important one around duplicate cookie headers (super specific to our setup, but could happen to anyone). We rolled back to master to restore login for a small set of users, fixed the issue, and are rolling the .NET Core branch again shortly.
Nope - that approach had other issues. I will say Set-Cookie has been one of the most painful parts of the migration.
.NET Full Framework did far too much magic (to support IE 5) which we had to design around, while .NET Core swings entirely the other way: it's all totally raw.
.NET Full Framework did far too much magic (to support IE 5) which we had to design around, while .NET Core swings entirely the other way: it's all totally raw.
That's it for today, we'll let master sit overnight and resume with alternative approaches to fix the duplicate cookie header issue tomorrow.
It's too deep to put into tweets really, but we'll be blogging about what we hit here in the migration.
Dark mode stays on though :)
It's too deep to put into tweets really, but we'll be blogging about what we hit here in the migration.
Dark mode stays on though :)
Alrighty - I'm mostly out with kids Tues/Thurs (our "new normal" around here), but the team is doing another full prod run of .NET Core today. Some people asked for some graphs so...why not.
Here's Stack Overflow on .NET Core, you can see when we deployed. CPU is down ~80%.
Here's Stack Overflow on .NET Core, you can see when we deployed. CPU is down ~80%.
The median page render time for questions dropped from about 21 ms (we were up a bit lately due to GC) to ~15ms.
The 95th percentile dropped from ~40ms to ~30ms (same measurement). 99th dropped from ~60ms to ~45ms.
Not too shabby, given we haven't optimized anything at all yet.
The 95th percentile dropped from ~40ms to ~30ms (same measurement). 99th dropped from ~60ms to ~45ms.
Not too shabby, given we haven't optimized anything at all yet.
We are *just porting* from .NET Full Framework 4.8 & ASP.NET MVC 5 to .NET Core 3.1.2 here.
Optimizations come later.
What we can see much more clearly is our own allocations (used to be hidden by so many MVC allocations). Plans to drastically reduce those soon.
Optimizations come later.
What we can see much more clearly is our own allocations (used to be hidden by so many MVC allocations). Plans to drastically reduce those soon.
Accordingly, memory isn't much different right now - there's a lot less gen0 noise but once we're into decent application lifetime, it's all "crap we have in memory" - no different between Full Framework and Core. But stay tuned, @marcgravell is working on the biggest allocation.
