I observed a couple of problems with the generator in some recent tests:
- there were a couple of hybrid test cases which did not have any
legacy nodes (randomness and all.) I change the probability to
produce more reliable results.
- added options to the generation to be able to add a max (to
compliment the earlier min) number of nodes for local testing.
- added an option to support reversing the sort order so "more
complex" networks were first, as well as tweaked some of the point
values.
- this refactored the generators cli parsing to be a bit more clear.
The previous implemention of hybrid set testing, which was entirely my
own creation, was a bit peculiar, and I think this probably clears thins up.
The previous implementation had far fewer legacy nodes in hybrid
networks, *and* also for some reason that I can't quite explain,
caused a test case to fail.
This changes the focus of the e2e suite, to (roughly) focus on
configurations that are more well used. Most production users of
tendermint run ABCI application in process and the GRPC/socket methods
cover the vast majority of the remaining use cases.
Perhaps we should consider drop support unix domain sockets in a
future release, but I think in the mean time it's useful to have the
tests *mostly* focus on the primary use cases.
When revwing #6807 I assumed that `probSetChoice` worked this way.
I think that the coverage of various configuration options should
generally track what we expect the actual useage to be to focus the
most test coverage on the configurations that are the most prevelent.
This is just a configuration change to default to using the new stack
unless explicitly disabled (e.g. `UseLegacy`) this renames the
configuration value and makes the configuration logic more clear.
The legacy option is good to retain as a fallback if the new stack has
issues operationally, but we should make sure that most of the time
we're using the new stack.
Having looked at our network address parsing and connection code, it
really looks like we're not doing anything on top of what the standard
library is doing (both in terms using `net.ParseIP` and also
`net.Dial`,) and I don't think we need to run the tests 2x the number
of times just to run through different areas of the standard
library. I think most of our users are going to be using IPv4, and
would be down to fully remove this dimension as well, if we find it's
making noise, but for now I think it's fine.
* Don't use state sync for nodes starting at initial height.
* Also remove stopped containers when cleaning up.
* Start nodes in order of startAt, mode, name to avoid full nodes starting before their seeds.
* Tweak network waiting to avoid halts caused by validator changes and perturbations.
* Disable most tests for seed nodes, which aren't always able to join consensus.
* Disable `blockchain/v2` due to known bugs.
In #5488 the E2E testnet generator changed to setting explicit `StartAt` heights for initial nodes. This broke the runner, which expected all initial nodes to have `StartAt: 0`, as well as validator set scheduling in the generator. Testnet loading now normalizes initial nodes to have `StartAt: 0`.
This also tweaks waiting for misbehavior heights to only use an additional wait if there actually is any misbehavior in the testnet, and to output information when waiting.
Closes#5291. Adds a randomized testnet generator. Nightly CI job will be submitted separately. A few of the testnets can be a bit flaky, even after disabling known-faulty behavior and making minor tweaks, and the larger networks may be too resource-intensive to run in CI - this will be optimized separately.
Sam Kleinman
Callum Waters
Aleksandr Bezobchuk
Marko
Erik Grinaker
Anton Kaliaev