~launchpad-pqm/launchpad/devel : changes

~launchpad-pqm/launchpad/devel » Changes from revision 11149.12.26

Rev	Summary	Authors	Date
11149.12.2...	Still not quite working right, but getting better. Still not quite working right, but getting better.	John Arbash Meinel	14 years ago
11149.12.2...	Trying to spawn multiple sessions seems to show some failure... Trying to spawn multiple sessions seems to show some failures. Need to investigate.	John Arbash Meinel	14 years ago
11149.12.2...	it only takes 907ms to start and stop the service, we may ne... it only takes 907ms to start and stop the service, we may need to look closer at what it takes to actually do a fork() then.	John Arbash Meinel	14 years ago
11149.12.2...	Create a command that just replays content that it read back... Create a command that just replays content that it read back to its output. That ensures that we've hooked up the various channels correctly.	John Arbash Meinel	14 years ago
11149.12.2...	rocks is too basic a test to be worthy of 2s runtime. rocks is too basic a test to be worthy of 2s runtime.	John Arbash Meinel	14 years ago
11149.12.2...	Unfortunately 'bzr rocks' test still takes 2s, though lp-ser... Unfortunately 'bzr rocks' test still takes 2s, though lp-serve takes 3.6s. However, things actually seem hooked up correctly now.	John Arbash Meinel	14 years ago
11149.12.2...	Change the code to allow --no-preload to help the test suite... Change the code to allow --no-preload to help the test suite.	John Arbash Meinel	14 years ago
11149.12.1...	I think I have it hooked up, but the test is failing. I think I have it hooked up, but the test is failing. Next step is to make sure the spawned process is using the correct log file for the test suite, so we get forking-service output in the test failure log file.	John Arbash Meinel	14 years ago
11149.12.1...	A command string => argv parser A command string => argv parser	John Arbash Meinel	14 years ago
11149.12.1...	change the name a bit, prepare for changing the 'fork' reque... change the name a bit, prepare for changing the 'fork' request. Instead of hard-coding what we fork and run, why not have it run a user-requested command. That should make it much more useful. Actually pretty similar to my old 'bzr-service' plugin.	John Arbash Meinel	14 years ago
11149.12.1...	Fix the test suite loader to actually load the tests. Fix the test suite loader to actually load the tests. It all looks like it is working again.	John Arbash Meinel	14 years ago
11149.12.1...	Merge in lp:devel, update lots of dependencies, finally get ... Merge in lp:devel, update lots of dependencies, finally get the test suite importing again.	John Arbash Meinel	14 years ago
11149.12.1...	try to move the test subprocess code into a common class. try to move the test subprocess code into a common class. At the same time, I now need to merge the latest launchpad devel code, because I just upgraded my system dependencies, and that seems to have broken some imports, etc. (The latest builder doesn't work with the current Person model, but the newer builder doesn't work properly with bzr-builddeb, etc, etc. I guess I shouldn't trust to simply update launchpad-dependencies without expecting to update my branch.	John Arbash Meinel	14 years ago
11149.12.1...	Add a test for the 'hello' heartbeat request. Add a test for the 'hello' heartbeat request.	John Arbash Meinel	14 years ago
11149.12.1...	Do lots of dancing to get it to work. Do lots of dancing to get it to work. It looks like ultimately we'll be back to running only in the zope test runner. The issue is that you have to use lots of 'canonical.' and 'lp.' code to determine where the 'bzr' and the 'python' scripts you need are. Which is duplicated with the code in 'codehosting/test_lpserve.py'. But for now, I'll live with this, because the tests are still faster. The one fork test really is slow, at about 4.3s to run. However, it already caught a bug that I was re-opening stdout for stderr, and never opening the stderr fifo. So it did catch something important.	John Arbash Meinel	14 years ago
11149.12.1...	Start trying to get some subprocess style testing for the se... Start trying to get some subprocess style testing for the service as well. I'm running into problems because of path issues. Specifically I run with a relative-to-cwd path to 'bzr' and TestCaseWithTransport does a chdir, but sys.argv seems to show only the relative path. So it cannot find the subprocess. Running with an absolute path to bzr fixes that problem, but I have the feeling it will confuse the zope test runner. So I'll need a different fix for that. After working around that, the next failure is because the BZR_PLUGIN_PATH isn't set correctly. I don't have any quick answers here yet.	John Arbash Meinel	14 years ago
11149.12.1...	Start building the testing infrastructure. Start building the testing infrastructure. Move lpserve into a plugin module rather than a simple script. The main benefit is that I can now invoke the test suite directly, rather than having to use the launchpad test runner, which is much heavier weight. (It takes >10s just to load the tests to filter it down to the test I want to run, I can run all tests under bzr selftest is 1.4s) Create a subclass for testing LPServe in the test suite, that makes it a bit easier to make sure that the class is started and stopped appropriately.	John Arbash Meinel	14 years ago
11149.12.9	add some tracing, fix some bugs add some tracing, fix some bugs I was using the file object, which select is okay with, but it was returning the file-id itself, and we were then mapping using the file and not the file-id, which was bad. I'm not getting a POLLHUP for the fifos, so I'm not really sure how to know when the remote service has shut down. It won't matter in practice, because we won't be using a poller, like we're doing here (twisted will be doing that). I've been trying to do some timing tests. As near as I can tell, after connect_to_* has started up, it is able to connect, send the request, and get the hello response in about 80ms. This is compared to 900ms to run 'bzr serve --inet', and 2400ms for lp-serve.	John Arbash Meinel	14 years ago
11149.12.8	Bring in the code from the other fork, but rework it. Bring in the code from the other fork, but rework it. The new TrivialForwarder is reworked, and split out into a simple script. We don't want to load all of bzrlib and lplib just to connect to the lp service (that is the whole point).	John Arbash Meinel	14 years ago
11149.12.7	Start working on a trivial client-side command. Start working on a trivial client-side command. I'd like to to perf test it, to make sure we actually start faster.	John Arbash Meinel	14 years ago

Older »