++1 from my side. It is very interesting how overall RunAll run time will be decreased.
My previous experience with TeamCity shows that removing pressure from the console and using adequate logging instead may bring x1.5 - x2 performance boost for tests. One more benefit of moving from synced streams to async logging will show us that bugs, which we can't see right now. So I absolutely agree to move logging data to a logger. пн, 11 февр. 2019 г. в 11:14, Maksim Stepachev <maksim.stepac...@gmail.com>: > Igniters, > > > When I was working with flaky tests, I was surprised that one of the > reasons for failure was a log appender blocking the console. > > > I suppose it happened because of a TeamCity agent communicating with java > out through Linux pipe. This is a problem for tests with 1GB log history. > > > Our test by default writes logs into 2 sources, such as the console and the > file. I'm going to change mode for the console at IGNITE_QUIET=true and > write only warn and error logs in it. Also, I’ll include a diagnostic log > into it. > > > You will be able to read the previous log from the Artifacts tab in your > build. > > This is a solution which has advantages: > > 1. First of all, we’ll cut down the space usage pre-test run by ~6-7 times. > For example - it's 1GB of saved space for 1 suit like "Cache (Restarts) 1". > > 2. We’ll be able to use a new space for long history storage for builds. > > 3. The TeamCity bot won't be lagging when it parses a large file. > > 4. Possibly some of tests won't be flaky. > > 5. And finally, the time of running my suite will be cut down by 5-7 > minutes. > > > Also, I'm going to make a workaround for previous behavior. For example, > this flag will be added to build params. But I should investigate it. >
