[dsmiley]

🤖 TestSolrCLIRunExample.testSolrCloudExampleWithScriptInputs failed on a busy CI worker with a misleading "collection not found" error, and the cause was in the tool, not only the test.

Failure chain

1. The test picks a free port by opening and closing a ServerSocket(0).
2. Before RunExampleTool probes it, another process takes that port. This is a race that only shows on a busy machine, such as a 96-core CI worker.
3. The tool prints "port N is already being used" and re-prompts.
4. --script-inputs is a fixed ordered list, so the retry consumes the next scripted answers (the collection name, then numShards) as ports.
5. isPortAvailable(2) hits the test security policy's SocketPermission denial, the tool aborts with exit 1, and Solr never starts.
6. The test ignored the exit code, so it failed later with "collection not found", hiding the real error.

Changes

• RunExampleTool: with --script-inputs, a port that is already in use now throws immediately with a clear message. A retry can't recover, since it would read the next scripted value.
• TestSolrCLIRunExample: testSolrCloudExampleWithScriptInputs asserts the tool's exit code, so a failure shows the tool's error. A new testScriptInputsPortInUse holds a port open and expects exit code 1.

Not changed
The test still picks a port, releases it, and uses it later, so it can still lose that race. It now fails with a clear message. The failExecuteScript access-denied output in the same log comes from the unrelated testFailExecuteScript, which expects it.

Testing
gradlew :solr:core:test --tests "org.apache.solr.cli.TestSolrCLIRunExample" passes (7 tests).

Related

• SOLR-18234: Reorganize org.apache.solr.cli into domain-based sub-packages #4663 (moves RunExampleTool and the test into sub-packages) and SOLR-18521: Unify picocli connection options into one ConnectionOptions mixin #5063 (touches RunExampleTool) overlap these files, so whichever merges second needs a rebase.
• SOLR-8545 (open, "TestSolrCLIRunExample is failing a lot on the jenkins cluster") may be related.

🤖 Generated with Claude Code

[epugh]

LGTM.

[epugh]

That's gnarly bug. Though the fix is good, both for making CI easier but also locally. I've been bit by running a script not realizing I had a zombied solr on port 8984 or was already running it in docker on 8983... and then getting my scripted run blows up obsucrely...