optimize(udp)/fix(quicSniffer): optimize performance of udp and fix a potential panic of quic #301
Conversation
![dae-prow[bot]](https://avatars.githubusercontent.com/in/345419?s=60&v=4){kind=small}
|
Hold it. We need to improve the stream sniffer instead of just skip the problem. I'll convert it into a draft. |
|
❌ Your branch is currently out-of-sync to main. No worry, I will fix it for you. |
mzz2017
changed the title
|
Ready for review. |
|
❌ Your branch is currently out-of-sync to main. No worry, I will fix it for you. |
Sniffing should work fine now. Please test it again, thanks. |
|
❌ Your branch is currently out-of-sync to main. No worry, I will fix it for you. |
|
TCP is working as expected. However, there is a noticeable speed discrepancy in UDP connection. Let’s gather more feedbacks from the end-users. |
|
https://t.me/daeuniverse/119500 It is recommended to merge in advance. |
|
cc @yqlbu |
Ok, let's do so. However, we shall make an announcement in the next release as this is a known issue. |
|
Link to #355 |
… potential panic of quic (daeuniverse#301)
Background
Fix panic
When a quic first packet is insufficient, the re-assembler will panic due to being out of range. This PR fixes it.
Support cross-packet reassembling for QUIC (SNI Sniff)
After this quic cross-packet reassembling, we can get SNI in more cases. However, quic connecting to google servers seems to be refused if the destination is overridden. Therefore, we can only use the quic sniffed domain for routing instead of dest overriding.
UDP Performance
And this PR improves UDP performance.
Before:
After:
Checklist
Full Changelogs
Issue Reference
Closes #343
Test Result