Occasionally we are seeing the go-test-api job timeout at 10 minutes.
Looking at the stack trace I saw the following:
1. Lots of tests blocked on server.Stop in NewTestServerConfigT. This
suggests that SIGINT is being sent to the server, but the server is
not properly shutting down.
2. Over 20k goroutines that look like this:
goroutine 16355 [select, 8 minutes]:
net/http.(*persistConn).readLoop(0xc004270240)
/usr/local/go/src/net/http/transport.go:2099 +0x99e
created by net/http.(*Transport).dialConn
/usr/local/go/src/net/http/transport.go:1647 +0xc56
Issue 1 seems to be the main problem, but debugging that directly is not
possible because our buffered logs do not get sent when the tests
timeout. To mitigate this problem I've added a timeout to the cmd.Wait()
to force kill the process and return an error.
Unfortunately because we retry this operation, we still may not see the
cause because the next attempt will likely pass. I'm tempted to remove
the retry around NewTestServerConfigT.
Issue 2 seems to be caused by not closing the response body. Since the
request is performed many times in a loop, many goroutines are created
and are not closed until the response body is closed.
Replaces #7559
Running tests in parallel, with background goroutines, results in test output not being associated with the correct test. `go test` does not make any guarantees about output from goroutines being attributed to the correct test case.
Attaching log output from background goroutines also cause data races. If the goroutine outlives the test, it will race with the test being marked done. Previously this was noticed as a panic when logging, but with the race detector enabled it is shown as a data race.
The previous solution did not address the problem of correct test attribution because test output could still be hidden when it was associated with a test that did not fail. You would have to look at all of the log output to find the relevant lines. It also made debugging test failures more difficult because each log line was very long.
This commit attempts a new approach. Instead of printing all the logs, only print when a test fails. This should work well when there are a small number of failures, but may not work well when there are many test failures at the same time. In those cases the failures are unlikely a result of a specific test, and the log output is likely less useful.
All of the logs are printed from the test goroutine, so they should be associated with the correct test.
Also removes some test helpers that were not used, or only had a single caller. Packages which expose many functions with similar names can be difficult to use correctly.
Related:
https://github.com/golang/go/issues/38458 (may be fixed in go1.15)
https://github.com/golang/go/issues/38382#issuecomment-612940030
Switch from /v1/agent/self to /v1/status/leader when checking if the test server has come up successfully in the waitForAPI function.
Previously, the test server was relying (probably not intentionally) on the default value of the acl_enforce_version_8 in the TestConfig, which was false. So if you create a test server and enabled ACLs, they would not be enforced and the server would be able to come up pretty quickly because /v1/agent/self would return a 200 status pretty much as soon as the agent is running and most likely before leader election is finished.
Now that we have removed acl_enforce_version_8 property (equivalent to being true by default) if you've created a test server with ACLs enabled, it will need to wait for leader election and for ACLs to be initialized before it'll get a successful response from the /v1/agent/self.
Note: With this change, waitForAPI function no longer requires a 200 response status from the v1/status/leader endpoint. This is because in some tests, namely TestAPI_AgentLeave, we are only running clients, and this endpoint returns a 500 status.
* Unflake the TestAPI_AgentConnectCALeaf test
* Modify the WaitForActiveCARoot to actually verify that at least one root exists
Also verify that the active root id field is set
* Fix zombie consul process in Windows
Windows doesn't support Interrupt signal, thus while stop it on Windows platform
it would fail and left zombie consul process
Also needed to update some funcs that were taking a *testing.T to use a testing.TB. This prevents passing a nil pointer as a non-nil interface value
and thus making it impossible to detect nil before using the interfaces functions.
This should cut down on test flakiness.
Problems handled:
- If you had enough parallel test cases running, the former circular
approach to handling the port block could hand out the same port to
multiple cases before they each had a chance to bind them, leading to
one of the two tests to fail.
- The freeport library would allocate out of the ephemeral port range.
This has been corrected for Linux (which should cover CI).
- The library now waits until a formerly-in-use port is verified to be
free before putting it back into circulation.
* Retry the creation of the test server three times.
* Reduce the retry timeout for the API wait to 2 seconds, opting to fail faster and start over.
* Remove wait for leader from server creation. This wait can be added on a test by test basis now that the function is being exported.
* Remove wait for anti-entropy sync. This is built into the existing WaitForSerfCheck func, so that can be used if the anti-entropy wait is needed
