mirrors/arti.git/tests, branch arti-v1.4.4mirror of https://gitlab.torproject.org/tpo/core/arti
http://git.dilluti0n.com/mirrors/arti.git/atom?h=arti-v1.4.42025-05-06T14:14:12ZMerge branch 'debug-ci' into 'main'2025-05-06T14:14:12Zgabi-250gabi@torproject.org2025-05-06T14:14:12Zurn:sha1:23dd43aa8fdfd0a98980bb15219369126827c64d
shadow CI tweaks
See merge request tpo/core/arti!2989shadow CI: enable model-unblocked-syscall-latency2025-05-05T22:15:51ZJim Newsomejnewsome@torproject.org2025-05-05T16:57:05Zurn:sha1:36969815dfa5ee6d09f3be483f61fd389c7ca73e
Not expected to solve any immediate issue, but helpful for consistency
with the chutney-shadow CI, and I think in the past we've hit busy-loop
deadlock without it if/when we try increasing the simulation end time.
Similarly, we apply the same tuning of "max_unapplied_cpu_latency" as we
use in the chutney-shadow simulation, which gets back some of the
simulation stability that enabling model-unblocked-syscall-latency
otherwise sometimes sacrifices.
shadow CI test: only require 1 successful HS transfer2025-05-05T22:08:22ZJim Newsomejnewsome@torproject.org2025-05-05T22:08:22Zurn:sha1:495b405cbb388ca6bbd6821154a1211184a10eda
As per previous discussion there is some known flakiness around hidden
services with TestingTorNetwork enabled:
<https://gitlab.torproject.org/tpo/core/arti/-/merge_requests/1399#note_2921505>
Previously, only requiring 5 of 10 transfers to succeed has been enough
to keep this stable, but recently we've gotten unlucky and dipped under
this threshold: <https://gitlab.torproject.org/tpo/core/arti/-/issues/1986>
More investigation is warranted, especially wrt why the problem seems
to mostly be affecting articlient-onion-auth, but for now it's better to
have a low success threshold than to have folks overriding the CI
failure.
shadow CI: disable strace logging2025-05-05T16:56:11ZJim Newsomejnewsome@torproject.org2025-05-05T16:56:11Zurn:sha1:dd5d3d76fba30fb1561c814907675253f3ebf261
While useful, it significantly slows down the test. We should probably
just enable it when needed.
integration-e2e-shadow: move simulation seed to yaml2025-04-23T15:34:55ZJim Newsomejnewsome@torproject.org2025-04-23T11:08:46Zurn:sha1:0d579bd48b73c963b41cba6c7529a98bd8b8c43c
Since this significantly affects the behavior of the simulation, it's
probably worth having it in the yaml. (It can of course still be
overridden from the command-line).
integration-e2e-shadow: move unblocked-syscall-latency to yaml2025-04-23T15:34:55ZJim Newsomejnewsome@torproject.org2025-04-23T11:04:57Zurn:sha1:3d82b403456d1559cfbcb604f37aede38899795f
Since this significantly affects the behavior of the simulation, it's
useful to have it in the yaml for reference or if the simulation is
manually rerun from the yaml.
integration-e2e-shadow: use a larger max_unapplied_cpu_latency2025-04-23T15:34:55ZJim Newsomejnewsome@torproject.org2025-04-22T21:28:53Zurn:sha1:46d4b9fe8889729a47adf2f61c7e6c5ba9fbba84
This should make the simulation results generally more stable with
respect to small perturbations, such as adding logging.
It also causes a previous "heisenbug" failure in this test to reliably
reproduce in every run.
tests: remove 'cargo build' from chutney setup2025-04-17T15:39:55ZSteven Engleropara@torproject.org2025-04-17T15:39:55Zurn:sha1:976d79f027525524925f889bd6e31aeb3da05af9shadow: capture only 40 bytes of packets2025-04-01T18:36:12ZSteven Engleropara@torproject.org2024-08-21T19:06:55Zurn:sha1:4cde5f568b01c972a130bf8f22b6f54ef5ccc4ec
The pcaps are only used to check for leaks by reading the packet
source/destination, so we don't need the entire packet. This should make
CI artifacts a bit smaller.
The fixed-size portion of an IP header is 20 bytes for IPv4 and 40 bytes
for IPv6. Shadow doesn't support IPv6, but used the IPv6 header size
anyways.
tor-circmgr: removed "ntor_v3" feature flag2025-04-01T14:17:21ZSteven Engleropara@torproject.org2025-04-01T00:56:00Zurn:sha1:c8fac43195c2baba6ab5464d2ea9e11dc0b0a141
ntor v3 is now always enabled.