search

tgstation / rust-g

Rust based libraries for tgstation
MIT License
28 stars 100 forks source link

Revert "3.0.0" #148

Closed MrStonedOne closed 10 months ago

MrStonedOne commented 1 year ago

Reverts tgstation/rust-g#147

Makes the server crash on round restart on windows.

This kills pop counts because people don't always reconnect and its common for players to decide to end a play session if they meet the slightest resistance.

optimumtact commented 1 year ago

do you have a stack trace?

MrStonedOne commented 1 year ago

No, stack traces go to stdout/stderr which only tgsv5 will save to disk and the only tgsv5 server is on linux which doesn't have this issue.

optimumtact commented 1 year ago

Then how do you know it's related to rust-g?

optimumtact commented 1 year ago

i mean the stack trace from the minidump

MrStonedOne commented 1 year ago

the timing matches up with the first 3.0.0 test merges and the fault offset shows up as rustg

optimumtact commented 1 year ago

they were crashing on the 25th and earlier too though, in fact you can see crashes on reboots all through july.

MrStonedOne commented 1 year ago

yes, randomly, not consistently. the crash rate goes from every few rounds to every round for sybil on the 18th, which is when the first test merge happened on sybil.

optimumtact commented 1 year ago

If I had to guess it's something to do with native mysql rust-tls probably using more tokio and causing crashes on exit.

MrStonedOne commented 1 year ago

it's probably just the generic 'all crashes are blamed on rust-g due to open connections'

ya if you bothered to read the pr comments you'd know that its being blamed because the timing matches up.

also my concerns when i open revert fix prs are always generally about fixing master, i can't pr to the release tab sorry but master on this repo should at least try to be stable

ZeWaka commented 1 year ago

ya if you bothered to read the pr comments you'd know that its being blamed because the timing matches up.

what comments, how am i supposed to know? tgstation/tgstation/pull/77307 has nothing and tgstation/tgstation/pull/76663 has nothing, and neither are linked here

do you mean this PR? I'm just reinforcing what oranges said

optimumtact commented 1 year ago

this isn't how you fix versioned packages.

The right answer would be a 3.0.1 which changes the cargo build script to use nativetls again.

But the problem is there's no definitive cause for the issue and i can't get enough information from the dmp file to actually determine what the root cause is.

MrStonedOne commented 10 months ago

byond has fixed the reconnect topic bug so crashing during restart is no longer a concern for pop counts.

also this might be fixed.