Also, probably, the memory leak problem solved in 6af9a62.

we should release buffers, after we're done with them, but we didn't do that in trojan-uot and udp-dns.

又在瞎猜了
1 bug是新版本出现的
2 忘记release不会导致leak 它只是把buffer返还给复用池 就算不还的话会被gc回收的

also, this matches the pprof chart:
freedom-ReadMultiBuffer -> udp-dispatcher-handleInput-callback -> trojan-handleUDPPayload-NewDispatcher-WriteMultiBuffer

my guess is that the memory leak problem is solved, anyway, this is clearly forgotten-release-buffer and should be fixed.

@kastov I am waiting for your confirm

also we should Interrupt(c.link.Writer) instead of Close(c.link.Writer) in udp-dispatcher, because we may have some data in link-cache

@kastov test https://github.com/XTLS/Xray-core/tree/katsov-test

这些稀碎东西能不能在你自己的frok试 别推到主仓库来污染眼球 just-test都来了

also we should Interrupt(c.link.Writer) instead of Close(c.link.Writer) in udp-dispatcher, because we may have some data in link-cache

@kastov test https://github.com/XTLS/Xray-core/tree/katsov-test

working on it

Here is also fresh dumps. Current ram usage 5.4GB. Yesterday was 3.2GB.

heap+gor+alloc-23082025.zip

I will relaunch now with core from branch which you provided

@kastov katsov-test branch updated, please use the new one.

@kastov katsov-test branch updated, please use the new one.

oh, i already started previous one. i believe it is better to keep that one for at least 24 hours to see results

if possible use new one for 24 h.

@kastov katsov-test branch updated, please use the new one.

btw any ideas how we can stresstest core to see if there any leaks? current way of testing is just so slow... :D

Here is also fresh dumps. Current ram usage 5.4GB. Yesterday was 3.2GB.

[heap+gor+alloc-23082025.zip](https://github.com/user-attachments/files/21950177/heap%2Bgor%2Balloc-23082025.zip)

I will relaunch now with core from branch which you provided

Try this https://github.com/XTLS/Xray-core/actions/runs/17186238756
Combining two pprofs, I reverted the only most suspicious commit

Hi there! Ram usage: 2.8GB

From branch: https://github.com/XTLS/Xray-core/tree/3236d737f665f99d12ee8c74eb4b86e5a4be1c5d (without last 4 commits).

heap+alloc+gor+24.zip

Try this https://github.com/XTLS/Xray-core/actions/runs/17186238756 Combining two pprofs, I reverted the only most suspicious commit

Working on it

i talk about trojan

Trojan ？那没事了，有爱玩的就自求多福吧，我可不管

first, this PR should be merged: #5050 (comment)

second, If you give me one day, I'll fix the trojan-uot-problem for the new version.

third, @Fangliding why you disable default-inbound-tcp-keepAlive???

纠正一下

// Get selects an arbitrary item from the [Pool], removes it from the
// Pool, and returns it to the caller.

Get() 后不 Put() 会被 GC，只是没起到 pool 的作用

there is no need for one day, i find a easy way, i fix it soon.

first, this PR should be merged: #5050 (comment)

second, If you give me one day, I'll fix the trojan-uot-problem for the new version.

third, @Fangliding why you disable default-inbound-tcp-keepAlive???

因为参考的nginx没开 服务端就算不开应该还是有 connIdle 来管 不过Trojan uot好像没有正确实现 connIdle。。

However, the Trojan uot does not seem to implement connIdle correctly.

yes, this is the reason, and i fix it in: #5064

Also, Shadowsocks 2022 has this memory leak issue #3131 . Does this PR fix the problem in Shadowsocks as well?

Also, Shadowsocks 2022 has this memory leak issue #3131 . Does this PR fix the problem in Shadowsocks as well?

first, #5064, fix the trojan memory leak problem, this is just some refines and i found them during searching of main problem.

second, no, i just modify trojan code.

@RPRX

this PR need rebase, wait...

第 1500 个 commit 留给你

READY.