Fix broken tests on elements and various cleanups by cdecker · Pull Request #3568 · ElementsProject/lightning

cdecker · 2020-03-04T19:43:21Z

This was supposed to be a regular cleanup PR, but given that a couple of tests broke along the way for liquid-regtest, it took me a bit longer.

Create test directory: some tests, especially those testing tools, will not create the working directory themselves, so we create it when the fixture is created.
Strengthened a couple of tx creation tests, mainly minimizing the array allocation sizes and ensuring everything is set up correctly (subsumes chain specific modifications).
MIgrate away from deprecate pylightning dependency towards pyln-client
There is a lingering test failure in which we fail to remove the working directory. I added a printing of the file listing to see what files are causing that issue. It then rethrows the error so we don't just swallow it.

There is one more thing I need to find before un-drafting:

If running with DEVELOPER=1, TEST_DB_PROVIDER=postgres and VALGRIND=1 there is a memory leak when retrieving UTXOs for txprepare.

Changelog-None

cdecker · 2020-03-16T14:09:00Z

I'll have to revisit the memleak when running with postgres later, don't want to hold up the other cleanups since they are rather nice imho.

cdecker · 2020-03-17T17:47:18Z

Managed to fix two more regressions:

The memleak that was present in postgres due to unordered UTXOs was also a leak with sqlite3 but manifested only rarely. Now fixed.
A new test that broke because it expected mainnet fees on liquid. Parameterized using chainparams

This should now be complete and ready to be reviewed. Paging release captain @niftynei 😉

cdecker · 2020-03-17T18:00:16Z

Here are the results from running all the various configurations:

[{"['TEST_NETWORK=regtest', 'VALGRIND=0', 'PYTEST_PAR=10', 'TEST_DB_PROVIDER=postgres', 'DEVELOPER=1']": '0'},
 {"['TEST_NETWORK=regtest', 'VALGRIND=0', 'PYTEST_PAR=10', 'TEST_DB_PROVIDER=sqlite3', 'DEVELOPER=1']": '0'},
 {"['TEST_NETWORK=regtest', 'VALGRIND=1', 'PYTEST_PAR=10', 'TEST_DB_PROVIDER=postgres', 'DEVELOPER=1']": '0'},
 {"['TEST_NETWORK=regtest', 'VALGRIND=1', 'PYTEST_PAR=10', 'TEST_DB_PROVIDER=sqlite3', 'DEVELOPER=1']": '0'},
 {"['TEST_NETWORK=liquid-regtest', 'VALGRIND=0', 'PYTEST_PAR=10', 'TEST_DB_PROVIDER=postgres', 'DEVELOPER=1']": '0'},
 {"['TEST_NETWORK=liquid-regtest', 'VALGRIND=0', 'PYTEST_PAR=10', 'TEST_DB_PROVIDER=sqlite3', 'DEVELOPER=1']": '0'},
 {"['TEST_NETWORK=liquid-regtest', 'VALGRIND=1', 'PYTEST_PAR=10', 'TEST_DB_PROVIDER=postgres', 'DEVELOPER=1']": '0'},
 {"['TEST_NETWORK=liquid-regtest', 'VALGRIND=1', 'PYTEST_PAR=10', 'TEST_DB_PROVIDER=sqlite3', 'DEVELOPER=1']": '0'}]

Some tests may not spawn a node at all, so make sure that our assumption that the directory exists in the fixture cleanup is correct by creating the directory.

rustyrussell · 2020-03-22T22:38:36Z

Simple rebase: test conflicts.

darosior

Can't ACK the element stuff, but otherwise looks good to me.

Leaved a few questions and restarted Travis

darosior · 2020-03-23T13:09:34Z

 else
 # Explicitly hand DEVELOPER and VALGRIND so you can override on make cmd line.
-	PYTHONPATH=`pwd`/contrib/pyln-client:`pwd`/contrib/pyln-testing:`pwd`/contrib/pylightning:`pwd`/contrib/pyln-proto/:$(PYTHONPATH) TEST_DEBUG=1 DEVELOPER=$(DEVELOPER) VALGRIND=$(VALGRIND) $(PYTEST) tests/ $(PYTEST_OPTS)
+	PYTHONPATH=`pwd`/contrib/pyln-client:`pwd`/contrib/pyln-testing:`pwd`/contrib/pyln-proto/:$(PYTHONPATH) TEST_DEBUG=1 DEVELOPER=$(DEVELOPER) VALGRIND=$(VALGRIND) $(PYTEST) tests/ $(PYTEST_OPTS)


darosior · 2020-03-23T13:11:58Z

 	}

-	if (pos == -1) {
+	if (pos >= tx->wtx->num_outputs) {


It cannot be >, can it ?

You're right. For some reason I'm in the habit of being overly cautious in these cases (spurious increments somewhere), but I'll fixup in this case :-)

darosior · 2020-03-23T13:12:40Z

+
+		/* Make sure we have a place to stash the witness script in. */
+		if (tal_count(tx->output_witscripts) < pos + 1) {
+			fprintf(stderr, "XXX resizing array to %d to fit item at %d\n", pos + 1, pos);


Trailing debug line ?

darosior · 2020-03-23T13:20:06Z

        'fundee_feerate_per_kw': 27625,
        'close_initiated_by': 'funder',
-        'expected_close_fee': 20333
+        'expected_close_fee': 33533 if chainparams['elements'] else 20333


How is this calculated ? Asking for ~~a friend~~ #3592 where I wondered how much to bump it in functional tests (e.g. 616a432#diff-d03575dee065f2dc319cf0e8d39ebe95R548)

I usually just binary search until I find params that work on both sides (keeping as many parameters as possible fixed).

I have not found a good way to derive from first principles, other than reimplementing the logic in python, which'd just double the maintenance effort without a major upside.

darosior · 2020-03-23T13:29:26Z

 		if (maxheight != 0 &&
-		    (!u->blockheight || *u->blockheight > maxheight))
+		    (!u->blockheight || *u->blockheight > maxheight)) {
+			tal_free(u);


Doing so will unreserve the utxo, right ? If so, I think that's what I missed to fix #3591 in #3593 !

Not yet: we are in wallet_select so the coins have not been marked as selected yet (that happens 6 lines down). It's the destructor added to the tallocated array on line 361 that'd free them, but they were not being added yet (just skipped) before this fix.

That's a roundabout way of saying that to unmark selected UTXOs you need to tal_free the utxos array :-)

We roll the `elements_add_fee_output` function and the cropping of overallocated arrays into the `bitcoin_tx_finalize` function. This is supposed to be the final cleanup and compaction step before a tx can be sent to bitcoin or passed off to other daemons. This is the cleanup promised in ElementsProject#3491

`pylightning` is not much more than an alias for `pyln-client`, so this removes the need to install that as well just to run the tests.

For some reason we fail to remove the test directory in some cases. My hypothesis is that it is a daemon that is not completely shut down yet, and still writes to the directory. This commit intercepts the error, prints any files in the directory and re-raises the error. This should allow us to debug the reappears.

pytest captures the output by monkey patching out `sys.stdout`. This may conflict with our use of `sys.stdout` when configuring logging, resulting in the "Write to closed file" issue that is spamming the logs. By making the logging configuration a fixture hopefully we always use the correct stdout (after pytest has monkey-patched it).

Looking for specific feerates, but not adjusting the amounts involved doesn't work.

Postgres does not guarantee that the insertion order is the returned order, which leads us to skip outputs that have already been stolen onto the selected utxos set, but not added to it because it isn't confirmed. This may also happen with sqlite3 though it's a lot rarer in that case.

Constants once again.

rustyrussell · 2020-03-23T23:22:26Z

Ack c85988f

cdecker force-pushed the cleanups branch 3 times, most recently from 7bd62a4 to e1e2801 Compare March 9, 2020 14:34

cdecker marked this pull request as ready for review March 16, 2020 14:09

cdecker requested a review from niftynei March 16, 2020 14:09

cdecker added this to the 0.8.2 milestone Mar 16, 2020

cdecker force-pushed the cleanups branch 3 times, most recently from 8b5a055 to d63cf6c Compare March 17, 2020 17:44

pytest: Actually make sure that the direcory exists

b5a41f9

Some tests may not spawn a node at all, so make sure that our assumption that the directory exists in the fixture cleanup is correct by creating the directory.

rustyrussell force-pushed the cleanups branch from d63cf6c to 4fc86f3 Compare March 22, 2020 22:38

darosior approved these changes Mar 23, 2020

View reviewed changes

cdecker added 9 commits March 23, 2020 16:00

tx: Make elements_add_fee_output private

8549308

pyln: Migrate remaining uses of the deprecated pylightning module

9aeb501

`pylightning` is not much more than an alias for `pyln-client`, so this removes the need to install that as well just to run the tests.

pytest: Unbreak the test_feerate_spam test for elementsd

136ec7a

Looking for specific feerates, but not adjusting the amounts involved doesn't work.

pytest: Make test_pay_retry less flaky for liquid-regtest

378037c

pytest: Fix test_closing_fee regression in elements

c85988f

Constants once again.

cdecker force-pushed the cleanups branch from 4fc86f3 to c85988f Compare March 23, 2020 15:00

rustyrussell merged commit e99ab03 into ElementsProject:master Mar 23, 2020

Conversation

cdecker commented Mar 4, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cdecker commented Mar 16, 2020

Uh oh!

cdecker commented Mar 17, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cdecker commented Mar 17, 2020

Uh oh!

rustyrussell commented Mar 22, 2020

Uh oh!

darosior left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rustyrussell commented Mar 23, 2020

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

cdecker commented Mar 4, 2020 •

edited

Loading

cdecker commented Mar 17, 2020 •

edited

Loading