cmcfarlen opened a new pull request, #13996: URL: https://github.com/apache/trafficserver/pull/13996
Cherry-picks for the 10.2.2 release from the `For v10.2.2` column of the [ATS v10.2.x project](https://github.com/orgs/apache/projects/573). These include #13976, which fixes the two Uranium tests that fail on the 10.2.x Fedora 45 job (`test_connect_curl` and `test_tls_verify_override_sni`). 15 commits, in master merge order: | # | PR | master SHA | Note | |---|---|---|---| | 1 | #13960 | `6ce3e997` | Fix abort on HTTP/2 origin request bodies with stream priority on | | 2 | #13961 | `2ddcb62b` | End HTTP/2 origin requests left open when no body follows | | 3 | #13791 | `34841db9` | stale_response: write the whole intercept body up front | | 4 | #13965 | `f903fbda` | urtest: drop nc and telnet from the accept timeout test | | 5 | #13954 | `6bdd742d` | Keep the socket error when kqueue reports it on both filters | | 6 | #13952 | `37675978` | Fix a net thread spin when a plugin pauses the outbound start hook | | 7 | #13956 | `0c0b332e` | Log pqssr as 1 only when the origin resumes the session | | 8 | #13584 | `251fc870` | Fix unsafe certificate common name extraction | | 9 | #13691 | `f08c0c8a` | Normalize bare carriage returns in header line endings — *drift* | | 10 | #13976 | `e7c59781` | Fix two Uranium tests on Fedora 45 | | 11 | #13621 | `2d54fe52` | Reject invalid QPACK static indexes — *test drift* | | 12 | #13977 | `086757e7` | Wait for the asserted h3 access log entry | | 13 | #13959 | `1e1aabd5` | Fix flakey config_reload_full_smoke race on early root success | | 14 | #13885 | `94a5f3b8` | Fix connect.down.policy 4 null server_txn crash | | 15 | #13991 | `06366073` | txn_box: release the peer certificate after reading its fields | All are `MERGED`, `base=master`, single-parent squash merges, none labeled `Incompatible`, none already present on 10.2.x. The 13 picks not marked *drift* change the same files, with the same line counts, as their master commits. ### Not picked - **#13953** (Read during the partial blind origin TLS handshake) fixes a stall in the write-readiness wait that #13512 added to `HttpSM::state_raw_http_server_open`. 10.2.x does not have #13512, so it does not have the bug. - **#13972** only changes the abuse_shield test, and abuse_shield is not on 10.2.x. - **#13980** (Uranium test Python dependencies) is covered by its direct 10.2.x backport, #13988. ### Adaptations - **#13691**: master first fixed repeated CRs in #13595, which is not on 10.2.x, and #13691 then replaced that check. `MIME.cc` matches master at #13691. `test_Hdrs.cc` gets only the `bare_cr_line_tests` table and the two new parser tests. The table includes #13595's four-CR case. - **#13621**: the QPACK, XPACK and header changes applied cleanly. 10.2.x's `test_QPACK.cc` lacks the infrastructure that #13636 added (`wait_for_events`, `QPACKTestAccess`, an atomic event handler), so the test was adapted: - The `TestQUICStream` ownership fix, `TestQPACKFixture`, `TestQPACKEncoderStreamDriver`, `wait_for` and `wait_for_event` come over as in the commit. The handler's event becomes `std::atomic<int>` because a test thread polls it. - Both out-of-range static index tests are as in master. - The self-eviction test drives its encoder stream instructions through the fixture and driver instead of `QPACKTestAccess`. 10.2.x's `QPACK` has no test friend, and this avoids adding one to a release branch header. The fixture takes the dynamic table size, and holds the instructions so they live as long as the driver. ### Testing Fedora 44 container, `ci-fedora-quiche` preset (QUICHE on BoringSSL, experimental plugins on): build clean, ctest 185/186. `test_qpack` passes all 6 cases, including the 3 new ones, as do `test_proxy_hdrs` and `test_proxy_hdrs_xpack`. The one failure is `uri_signing_test`, on cjose HMAC `crypto error` against BoringSSL. None of these picks touch uri_signing or cjose. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
