Skip to content

Commit

Permalink
Use less subshells and tees in running tests with bazel run.
Browse files Browse the repository at this point in the history
Fixes #17754.

What we have seen prior to this change was that sometimes for quick tests the output was swallowed. After a lot of poking it became clear that the culprit is the use of subshell and `tee`, e.g. if you remove `tee` completely from the picture the behavior never shows up.

The issue is that with a fast test, `tee` seems to be killed (or its parent subshell) before the printing the output to stdout.

With this change, we reduce the number of subshells and processes to set up and reduce the chance of the race condition but not remove it.

However, for practical purposes, the race condition is gone.

With the reproduction steps in #17754, and this command
```
for i in {1..10000}; do /tmp/bazel run :foo &> /tmp/log ; grep -q "useful echo" /tmp/log ; if [ $? -eq 0 ]; then echo -n '+'; else  echo -n '-'; fi; done
```
a bazel from head fails ~3900 out of 10000 times.

After this commit, it never failed.

Closes #17846.

PiperOrigin-RevId: 518794237
Change-Id: I8c1862d3a274799b864f0f5f42b85d6df5af78c7
  • Loading branch information
meisterT authored and copybara-github committed Mar 23, 2023
1 parent c82168e commit c04f0d4
Showing 1 changed file with 2 additions and 2 deletions.
4 changes: 2 additions & 2 deletions tools/test/test-setup.sh
Original file line number Diff line number Diff line change
Expand Up @@ -323,9 +323,9 @@ if [[ "${EXPERIMENTAL_SPLIT_XML_GENERATION}" == "1" ]]; then
fi
else
if [ -z "$COVERAGE_DIR" ]; then
("${TEST_PATH}" "$@" 2> >(tee -a "${XML_OUTPUT_FILE}.log" >&2) 1> >(tee -a "${XML_OUTPUT_FILE}.log") 2>&1) <&0 &
("${TEST_PATH}" "$@" 2>&1 | tee -a "${XML_OUTPUT_FILE}.log") <&0 &
else
("$1" "$TEST_PATH" "${@:3}" 2> >(tee -a "${XML_OUTPUT_FILE}.log" >&2) 1> >(tee -a "${XML_OUTPUT_FILE}.log") 2>&1) <&0 &
("$1" "$TEST_PATH" "${@:3}" 2>&1 | tee -a "${XML_OUTPUT_FILE}.log") <&0 &
fi
fi
childPid=$!
Expand Down

0 comments on commit c04f0d4

Please sign in to comment.