fa272dfdff711e324baefc9bdbb2bf4b52d99eb0 ci: Properly bump to focal for win cross build (MarcoFalke)
Pull request description:
Fixes#21122
ACKs for top commit:
fanquake:
ACK fa272dfdff711e324baefc9bdbb2bf4b52d99eb0 - Looks good to me. I can see the wine version output, `wine-5.0 (Ubuntu 5.0-3ubuntu1)`, in the [log](https://cirrus-ci.com/task/5743559502462976), and `make check` is running.
Tree-SHA512: cd37462afc5512e00cef5e9e7fd1bb5c43c600e833b30cdea2c1c443dc7b0e68f5f2cbaf9d7b655892059af0a226478211db1dc97c10fe4dcdfe666a784f3afc
ac24af453d16afd1993fa1f292aa41ae6b16f138 ci: use Ubuntu Focal for macOS cross build (fanquake)
Pull request description:
I had assumed Cirrus was spinning up Docker containers to run the CI,
however we are actaully running on the Cirrus machines themselves. See
`DANGER_RUN_CI_ON_HOST` and in the logs:
```bash
Running on host system without docker wrapper
```
So with this change we will actually be using Focal for the macOS cross build.
Follow up to #21036.
This originally contained Windows changes, and an attempt to get Cirrus running without `DANGER_RUN_CI_ON_HOST`, however that seems non-trival, so Windows changes have been dropped from here for now.
ACKs for top commit:
MarcoFalke:
cr ACK ac24af453d16afd1993fa1f292aa41ae6b16f138
Tree-SHA512: 587ba5acf741bcefecf1bc262fa1177f565ebfa9de56125eca19ed3c7db7b9aabfb96866e9c140681b88cb7015a3ded2bc6b4b1b235543d6f6e9dfc6984d569f
faeb40bee3bca9477785659d89af53cafa2333b5 ci: Install missing lint packages (MarcoFalke)
Pull request description:
The cirrus container is vanilla ubuntu, so we need to install the needed packages
ACKs for top commit:
hebasto:
ACK faeb40bee3bca9477785659d89af53cafa2333b5, I have reviewed the code and it looks OK, I agree it can be merged.
Tree-SHA512: e56198108e26ea0ba2a344b1b74bc294652f34e9866cca053a25fb1b83bbd87ea40254c340e5e169fdfcbd4dcb39fdc2078b5157ca729a22a9a1792ec514a33e
95487b055328b590ba83f258de9637ab0f9a2f17 doc: Drop mentions of Travis CI as it is no longer used (Hennadii Stepanov)
09d105ef0f8b4b06bf248721a1209c9e16e9db75 ci: Drop travis_fold feature as Travis CI is no longer used (Hennadii Stepanov)
Pull request description:
As Travis CI is no longer used, this PR:
- drops `travis_fold` feature
- drops mentions of Travis CI in docs
ACKs for top commit:
MarcoFalke:
ACK 95487b055328b590ba83f258de9637ab0f9a2f17
Tree-SHA512: 2e259bb8b1e37bcefc1251737bb2716f06ddb57c490010b373825c4e70f42ca38efae69a2f63f21f577d7cee3725b94097bdddbd313f8ebf499281cf97c53cef
facf5e37f619be6bd7f8768058842679cd9ad933 ci: Only use credits for pull requests to the main repo (MarcoFalke)
Pull request description:
No need to spend credits for the `master` branch, because the build shouldn't fail there anyway and it is not time-critical to get a fast feedback.
Some other changes:
* Disable `stateful` for faster scheduling
* Reduce lint memory from 8G to 1G for faster scheduling
ACKs for top commit:
hebasto:
ACK facf5e37f619be6bd7f8768058842679cd9ad933, I have reviewed the code and it looks OK, I agree it can be merged.
Tree-SHA512: 4d052e68217086574b9ea3d603cde1b585833c289d47dfed5308ff001d02964dc75ec3b3ebf5b233ccd09c47ad4ff5ba0bef639bf6362d984e7c49fca8fec24b
4045a6722c884be779e86016313061ac6ff80808 ci: Use cpu=1 for linter (Dhruv Mehta)
739d39022d2959c1114c14a0065daebf4fe812c1 ci: Move linter task to cirrus (Dhruv Mehta)
Pull request description:
Solves #20467: Move linter to Cirrus-CI as Travis-CI.org is shutting down
ACKs for top commit:
MarcoFalke:
ACK 4045a6722c884be779e86016313061ac6ff80808
Tree-SHA512: 9aa7487ac86c91fc68bb584d29134e304dbd46702514a5d47d1ef0de6b877d96d42b7589870fc67ad9a31f5d3a789728446da4418688f336111a9ba0f8de5feb
faf2c6e32e5b27cf936c9b26e6059027dc020374 cirrus: Schedule one task with paid credits for faster CI feedback (MarcoFalke)
Pull request description:
During times of high activity in the repo, the scheduling of Cirrus CI tasks might put them a few hours in the future. This is fine when all the tasks eventually pass. Though for failing tasks, a failure should ideally be shown to the author and reviewer as soon as possible.
Compute credits can be used to schedule immediately: https://cirrus-ci.org/pricing/#compute-credits. Running all tasks with compute credits will probably be more expensive than our previous CI invoice. However, they are also more flexible.
As a start we could enable only a single task and revisit/re-evaluate the next steps in a month.
ACKs for top commit:
laanwj:
Concept ACK faf2c6e32e5b27cf936c9b26e6059027dc020374
fanquake:
ACK faf2c6e32e5b27cf936c9b26e6059027dc020374
practicalswift:
cr ACK faf2c6e32e5b27cf936c9b26e6059027dc020374: patch looks correct
Tree-SHA512: df599e1c4cc0394f7f03413ad29954ddc87b163b02640d8bfc0497a5dc8d237b8c963c1dd9d01ac83c4a044f575300763097dac2ea6d1a4a163a1cece342b743
fa5c4f12f5d017d8dbe7700ca9688189eacbf32b ci: Adjust cirrus ci task names (MarcoFalke)
Pull request description:
The task names are too long for GitHub to display them properly without truncation in the "checks-view". Fix that by using a new naming scheme:
* Native builds don't mention "x86_64 Linux", as it is redundant, they do mention the OS (bionic or focal) in the name suffix
* Cross builds mention the target in the prefix and the OS (always bionic) in the suffix
* the macos native build simply says "macos native"
ACKs for top commit:
practicalswift:
ACK fa5c4f12f5d017d8dbe7700ca9688189eacbf32b: patch looks correct!
hebasto:
ACK fa5c4f12f5d017d8dbe7700ca9688189eacbf32b, I have reviewed the code and it looks OK, I agree it can be merged.
RiccardoMasutti:
ACK fa5c4f1
Tree-SHA512: 856deb0577c97c70069ef1d369991addc49522135c0ad9e382218fd79ba3d55a95d6c601288dcef0510764b92fbd30a9d7de32b08dc5be55482deab14049b892
2b356117e94f9ef27b67a8e98663f5d676f58c11 ci: no-longer exclude feature_block in TSAN job (fanquake)
Pull request description:
The TSAN job is now running on Cirrus.
Increase the allocated memory to the [maximum allowed](https://cirrus-ci.org/guide/linux/#linux-containers).
ACKs for top commit:
jonasschnelli:
utACK 2b356117e94f9ef27b67a8e98663f5d676f58c11 - checked the CI run and confirmed that the feature_block runs: https://cirrus-ci.com/task/6008403543719936?command=ci#L3249
MarcoFalke:
review ACK 2b356117e94f9ef27b67a8e98663f5d676f58c11
Tree-SHA512: b774995600361c74bc3267b566e12add66a4604bdf88f6e3f69669edbb8d7aff6f20fdbf0ef98187be4730ce4e18b1939bbcecd993a5c5c1ff40b237c7921b71
fa7a4385d08797876de9a05ae8a6fde18e36bd67 ci: Fix doc typos in .cirrus.yml (MarcoFalke)
fa73674738b0b5a271ae387e2ebd76d747e02d91 ci: Run i686 centos ci config on cirrus (MarcoFalke)
fa1f949a4dc2568456f3ba30bbd8ecbc78a423ca ci: Run nowallet ci config on cirrus (MarcoFalke)
Pull request description:
Travis CI Org is shutting down, so move the configs to cirrus ci
ACKs for top commit:
practicalswift:
ACK fa7a4385d08797876de9a05ae8a6fde18e36bd67: patch looks correct
Tree-SHA512: 1b7125c7f0d2288931fb8c5e90345891d5f7c1f00c4af136afbeb36bd2836f72920a1877dacc7be787e2c8d84de2a733c1ca71931e5c101b222c1fea588619b3
fa76d8d4d71d844e217686881d4f630eac3a8e10 test: Actually print TSan tracebacks (MarcoFalke)
Pull request description:
Commit 5e5138a721738f47053d915e4c65f925838ad5b4 made the TSan logs to be printed before returning an error from the ci script.
However, it seems that on Cirrus CI, the `--failfast` option will kill not only all python process and bitcoind child process, but also the parent CI bash script, rendering the `trap` inefficient. I believe this bug was introduced in commit 451b96f7d2796d00eabaec56d831f9e9b1a569cc.
ACKs for top commit:
fanquake:
utACK fa76d8d4d71d844e217686881d4f630eac3a8e10
Tree-SHA512: 686f889d38a343882cb62ad6e0c2080196330e7cc7086891a7ff66d9443b455c82ba8d7e4a5cc42daa0513b0ad2743055bfe90e2f6ac88a910ee3b663fabddcd
fa35efa84bb614df9629076e45e9e974d9e940f6 ci: Run arm task on arm64 hardware (MarcoFalke)
Pull request description:
It will still run cross-compilation to armhf, but run the binaries on the hardware itself, not qemu.
There shouldn't be any significant difference, other than maybe a slight speedup.
ACKs for top commit:
hebasto:
re-ACK fa35efa84bb614df9629076e45e9e974d9e940f6
Tree-SHA512: d03bb2b55d0a6b5021243eedff6e143f4fad03a1a8913000c2c5c74665e4bb1c77cb30aa112dbddbc0bb7c645bec18f6b8b8ba91049587b464bbe5ec07dd46bd
20e491ddcb2617472c15294067768e8ce122499a CI/Cirrus: Skip merge_base step for non-PRs (Luke Dashjr)
Pull request description:
CIRRUS_BASE_BRANCH is a PR-specific variable and undocumented on non-PR builds.
In practice (at the moment), it seems to be HEAD, which in private repositories can be pretty much anything, causing CI to fail if it can't be cleanly merged.
By checking CIRRUS_PR first, we can reliably do CI builds of branches outside PRs.
ACKs for top commit:
MarcoFalke:
review ACK 20e491ddcb2617472c15294067768e8ce122499a
Tree-SHA512: 9fd8db2e19a3145f7dccfca107631b20df8c94d385f624e2bcef2fa18e38bf3e23c6c68fc8241decedbf1413bf69ca572cff75e1ccf82c09ac50443001ec5ae5
fa92cf29d971811ff8dd844ee1a1e2e12b79e389 ci: Remove redundant valgrind fuzz task (MarcoFalke)
Pull request description:
This task has several issues:
* It slows down other tasks and times out: It needs a lot of resources (CPU, RAM, time), because it builds more than 100 fuzzers, clones a 2 GB repo with 100k seeds and pipes them all through valgrind
* It doesn't add a lot of value: Except for one issue in the boost time library, it hasn't found any issues that the existing fuzz,asan,ubsan fuzzer has already found
* It is redundant: It is already run in the bitcoin-core/qa-assets repo on every push of new seeds and once daily
Fix all issues by removing it here.
Top commit has no ACKs.
Tree-SHA512: 76d16a3e5afc79ba9d89cfeb915d7e66fd1ad4e2035ae4ccd30a21cc060bfba3fb6a904346b7c41606e69d972e86bf660df962673028689227cc26072bba24bd
77777c8b5e3d0edbe1178d982a778b7cc9964b70 ci: Run windows ci config on cirrus (MarcoFalke)
3333d6942e8c5d823b97ea51ef88c8a0791742c6 ci: Run macos ci config on cirrus (MarcoFalke)
fa8b1114e60100f503673ccc7ef5618851c0c9e3 ci: Run arm ci config on cirrus (MarcoFalke)
fa0795f54dc5268bacc6acdfdbe23f9952a0e7d8 ci: Replace TRAVIS_OS_NAME with CI_OS_NAME (MarcoFalke)
fafce1a13a65cc3ae0cd5351648bb0290f9bfe8c ci: Move documentation to correct config file (MarcoFalke)
Pull request description:
ACKs for top commit:
hebasto:
ACK 77777c8b5e3d0edbe1178d982a778b7cc9964b70
Tree-SHA512: 581d3bdb2c3e3da20bd8492c6b23d90f3b4f7f9300ade384667e41de9fbbe42bbcef26dd99328839dc0b49be6e303c4c6d1b66ceb5d8009800c52c97685f9080
66667acc53cdfe224b36a74be4ba046b28282d15 cirrus: Skip tasks on the gui repo main branch (MarcoFalke)
Pull request description:
No need to run every build twice, once in the main repo and then in the read-only gui mirror repo
ACKs for top commit:
decryp2kanon:
ACK 66667ac
hebasto:
ACK 66667acc53cdfe224b36a74be4ba046b28282d15, though still preferring `only_if` as showing skipped tasks as successful ones seems a bit confused.
Tree-SHA512: 0d35bd115152e06ba4dc5f364130ba5496167d960c44eac2c76192ff9bf7c51f46ab72e2d054dcc6a91818a18dffbbc262f8a4c4483857158c0af4f55dfe9b28
fa8e494554d0f5f8f48bb33ce96a8d7de9b95e31 ci: Run ci configs on cirrus (MarcoFalke)
Pull request description:
Now that cirrus ci runs more stable than travis ci, we can try to move more configs over there to see if any issues arise.
ACKs for top commit:
practicalswift:
ACK fa8e494554d0f5f8f48bb33ce96a8d7de9b95e31: patch looks correct
decryp2kanon:
reACK fa8e494554d0f5f8f48bb33ce96a8d7de9b95e31
Tree-SHA512: e2d1838050b6199d11fa06d1cc9d804883ec5df7d65386c950e8124c0067dc1aaa62ec84c9842c8263e2cf5b17fc819ce85689338113f8d69edb1954f06e76e2
faf2999e2515c47108dc3d376dbd1c0fce4d6103 cirrus: Use kvm to avoid spurious CI failures in the default virtualization cluster (MarcoFalke)
Pull request description:
Try to fix#20093
ACKs for top commit:
practicalswift:
ACK faf2999e2515c47108dc3d376dbd1c0fce4d6103
hebasto:
ACK faf2999e2515c47108dc3d376dbd1c0fce4d6103, the related doc:
Tree-SHA512: 156aa2ce5a5dde11570f7f90f9d51be540a5469f090033ab6f337b2c46347741469e109f9566f4b7c424339483d61192b91102021f5db38823ce4cf4428e5671
fa8e1487144eab237ffd291397355ef4801f46f8 ci: Double tsan CPU and Memory to avoid global timeout (MarcoFalke)
Pull request description:
Fix#19864
ACKs for top commit:
practicalswift:
ACK fa8e1487144eab237ffd291397355ef4801f46f8 -- patch looks correct
hebasto:
ACK fa8e1487144eab237ffd291397355ef4801f46f8, according to https://cirrus-ci.org/guide/linux/ the limits are:
Tree-SHA512: b6d522290bfe80ed7453387b811628bf42c7657aa6a84d2f5984c8bb16f9857a71eabc6b8a4d63b84227d59b41a8ed7dd85d86cae5628dc9cf6b85bd365248d7
fa0538e94db26dd84e02aac1cf174b79729dae72 ci: Set cirrus RAM to 8GB (MarcoFalke)
fa41810d0e87f9f9a2e39be238b9598be02646d0 ci: Run valgrind fuzzer on cirrus (MarcoFalke)
Pull request description:
The first commit should fix the 50min timeout in forked repos. Similar to #19424. E.g. https://travis-ci.org/github/bitcoin-core/gui/builds/718322267
The second commit should fix#19744
Top commit has no ACKs.
Tree-SHA512: c765098dfa913ca49b1d1eee99aaa83e4b9eb191b7ad5e652e3f04744fe8670dd3ef4215832b8e2b5bac0273d24f607fc275e72f566326108ba42ab57228ffd4
fa8e6df282af0d396d75b03721f1b59a520ced19 ci: Run tsan ci config on cirrus (MarcoFalke)
Pull request description:
Fixes bitcoin-core/gui#12
Copied description from #19321:
Currently it is not possible to use travis in forked repositories due to the 50 minute limit on builds. A fresh build (uncached) of the thread sanitizer config takes more than 50 minutes.
One approach to fix this could be to throw away tests until the run time is less than 50 minutes. However, the risk of being blind of failures in the thrown away tests is not worth the gain. Also, to detect them, one has to run the tsan configuration nightly and failures could only be detected post-merge.
Another approach would be to ask travis support to raise the limit for a forked repository. This is a tedious and manual one-by-one process, so I'd rather not.
Finally, a different ci provider can be used, since the config files are designed to be platform-agnostic. This is what I picked.
I kept all settings identical to the travis machine for now. Both providers run in the google cloud, so this should be a "move-only".
ACKs for top commit:
fanquake:
ACK fa8e6df282af0d396d75b03721f1b59a520ced19 - my understanding is that test coverage remains the same. Just swapping providers to work-around the Travis time-limit in other repos.
Tree-SHA512: 26fed248a4f743107160d3b9e5df57fa0be280fd065ae6fece83d254f59d58ccf3e11a245519d158da109c47b053f62ee8756215008541973c65dc28c4efb748
fa2eb3d5d6819e42bfcec8a9f02b99438fe718b9 ci: Run asan ci config on cirrus (MarcoFalke)
fa93527738a62ebc13305adcb0fd2b5128073bbc cirrus: Clear dummy task (MarcoFalke)
Pull request description:
Currently it is not possible to use travis in forked repositories due to the 50 minute limit on builds. A fresh build (uncached) of the address sanitizer config takes more than 50 minutes.
One approach to fix this could be to throw away tests until the run time is less than 50 minutes. However, the risk of being blind of failures in the thrown away tests is not worth the gain. Also, to detect them, one has to run the asan configuration nightly and failures could only be detected post-merge.
Another approach would be to ask travis support to raise the limit for a forked repository. This is a tedious and manual one-by-one process, so I'd rather not.
Finally, a different ci provider can be used, since the config files are designed to be platform-agnostic. This is what I picked.
I kept all settings identical to the travis machine for now. Both providers run in the google cloud, so this should be a "move-only".
ACKs for top commit:
hebasto:
ACK fa2eb3d5d6819e42bfcec8a9f02b99438fe718b9
Tree-SHA512: 159d7dc6f5b24583e941282cdd40465b15db787f0a658a3e81a7b1a22abdb4cb573709b9b5c4465523e0ba0060b17a68fbdbda7a9ecdeb649f31535d377bbe75
BACKPORT NOTICE:
this PR doesn't actually swithc to libc++ due to multiple CI failures such as
linking errors or other
------------------
faf62e6ed0ca45db44c370844c3515eb5a8cda12 ci: Remove unused workaround (MarcoFalke)
fa7c8509153bfd2d5b4dcff86ad27dfd73e8788b ci: Install llvm to get llvm symbolizer (MarcoFalke)
fa563cef61e8a217c5e8ec059e174afae61087a5 test: Add more tsan suppressions (MarcoFalke)
fa0cc02c0a029133f080680ae9186002a144738f ci: Mute depends logs completely (MarcoFalke)
fa906bf2988c799765a04c484269f890964ec3ee test: Extend tsan suppressions for clang stdlib (MarcoFalke)
fa10d850790bbe52d948659bb1ebbb88fe718065 ci: Use libc++ instead of libstdc++ for tsan (MarcoFalke)
fa0d5ee1126a8cff9f30f863eb8f5c78bf57e168 ci: Set halt_on_error=1 for tsan (MarcoFalke)
fa2ffe87f794caa74f80c1c2d6e6067ee4849632 ci: Deduplicate DOCKER_EXEC (MarcoFalke)
fac2eeeb9d718bdb892eef9adf333ea61ba8f3d0 cirrus: Remove no longer needed install step (MarcoFalke)
Pull request description:
According to the [ThreadSanitizer docs](https://clang.llvm.org/docs/ThreadSanitizer.html#current-status):
> C++11 threading is supported with **llvm libc++**.
For example, the thread sanitizer build is currently not checking for double lock of mutexes.
Fixes (partially) https://github.com/bitcoin/bitcoin/issues/19038#issuecomment-632138003
ACKs for top commit:
practicalswift:
ACK faf62e6ed0ca45db44c370844c3515eb5a8cda12
fanquake:
ACK faf62e6ed0ca45db44c370844c3515eb5a8cda12
hebasto:
ACK faf62e6ed0ca45db44c370844c3515eb5a8cda12, maybe re-organize commits to modify suppressions in a single one?
Tree-SHA512: 98ce5154b4736dfb811ffdb6e6f63a7bc25fe50d3b73134404a8f3715ad53626c31f9c8132dbacf85de47b9409f1e17a4399e35f78b1da30b1577167ea2982ad
cbd661122e5852d543467090459d33cf8cb4a3c7 Set LD_LIBRARY_PATH consistently in travis tests (Russell Yanofsky)
fa35c34df781cf46bbd15522961f214f03b958bf Remove unused ci configs that have been moved elsewhere (MarcoFalke)
3333cb96994182bbdbb21174b691feb716858bc2 fuzz: Pass down MAKEJOBS to test_runner (MarcoFalke)
Pull request description:
Just how `MAKEJOBS` is passed down to the functional test `test_runner`, do the same for the fuzz `test_runner`.
Also includes a commit to remove unused config files, which have been moved elsewhere.
Top commit has no ACKs.
Tree-SHA512: 32557102c9e40599b432aeb004c8427e8fbb07cdf4048050cdc8241d1b029aaad306b1131007eeca8315a4f71c38a7efbb833310e056cd11b835676cd19b8902
6136a96cdf631f824fb89cf2720422787439acc9 ci: Rename RUN_CI_ON_HOST to DANGER_RUN_CI_ON_HOST (Hennadii Stepanov)
97ba77aa8e58c3c7ff767e6ca804d0e46370e5dc ci: Add native s390x (Hennadii Stepanov)
Pull request description:
Unlike the Docker wrapped solution (#17591) this PR suggests running on host system directly.
This approach makes builds quick and stable (see: #18106).
The excerpt from the Travis log:
```
...
Running on host system without docker wrapper
...
Byte Order: Big Endian
...
```
ACKs for top commit:
MarcoFalke:
ACK 6136a96cdf631f824fb89cf2720422787439acc9
Tree-SHA512: 1b591de13e38d10a35217e1de11cbd648a359d18d16eed166fac18ea5788b58cc9fc6d407086ed342b99e57e479efd951a0ea693710177e500eb116316b9a788
facc0da63a8fa4bd6fc2782cbe92eb9f920f2256 travis: Run unit and functional tests on native arm (MarcoFalke)
fafa064d2a8dbe24303545ab582ec84cde52ab5b ci: Remove ccache requirement on the host (MarcoFalke)
Pull request description:
This keeps the cross-compilation to make it easy to run the ci on non-arm hardware. To run this locally in qemu-user as it used to be, just `export QEMU_USER_CMD="qemu-arm -L /usr/arm-linux-gnueabihf/"`.
ACKs for top commit:
laanwj:
LGTM ACK facc0da63a8fa4bd6fc2782cbe92eb9f920f2256
practicalswift:
ACK facc0da63a8fa4bd6fc2782cbe92eb9f920f2256 -- diff looks correct and Travis seems happy
Tree-SHA512: 0dc1bc82eb93e2bd8b159e044f20fe3055f8cdfd73aaa238bd2e178397582144dfc0c6a87bd8270115dafea1a623e642bde5d5f30254f94140f1a2cdb12fc2da
fa69588537bc91c0aedbc89ef1760d89cbffad75 test: Make PORT_MIN in test runner configurable (MarcoFalke)
Pull request description:
This is needed when some ports in the port range are used by other processes. Note that simply assigning the ports dynamically does not work:
* We spin up several nodes per test (each node gets its own port)
* We run several tests in parallel
So to avoid nodes from different tests colliding on ports, the port assignment must be deterministic (can not be dynamic).
Fixes: #10869
ACKs for top commit:
practicalswift:
ACK fa69588537bc91c0aedbc89ef1760d89cbffad75 -- diff looks correct
promag:
ACK fa69588537bc91c0aedbc89ef1760d89cbffad75.
Tree-SHA512: e79adb015e7de79064e2d14336c38bc9672bd779ad6c52917721897e73f617c39d32c068a369c26670002a6c4ab95a71ef3a6878ebdd9710e02f410e2f7bcd14
fab98992043f47fa7240d7c1217920d0c4f783a2 test: Try once more when RPC connection fails on Windows (MarcoFalke)
faa655731eac751d4eb494268e2c815493ba9382 test: Document why connection is re-constructed on windows (MarcoFalke)
fa9f4f663c36b0824406036445e5cff0a78174e9 test: Remove python 3.4 workaround (MarcoFalke)
fae760f2b24cb26494b65c0a7ac38b92ead345af cirrus: Bump freebsd to 12.1 (MarcoFalke)
Pull request description:
Fixes: #18548
ACKs for top commit:
hebasto:
ACK fab98992043f47fa7240d7c1217920d0c4f783a2, I have reviewed the code and it looks OK, I agree it can be merged.
Tree-SHA512: c4e9ed8d995b63a820ca66984f152ac216c83ba1f318b61b15c6d375c0e936c08f6bc3d38c255dddf3ee8952f848c7ababf684854e07a7c1b1d8504e6b7208ba
