Ben Pfaff

@mihaibudiu I got a test failure running your script: ``` FAILURE 0.0sec, org.apache.calcite.test.JdbcFrontJdbcBackTest > executionError java.util.concurrent.RejectedExecutionException: Thread limit exceeded replacing blocked worker at java.base/java.util.concurrent.ForkJoinPool.tryCompensate(ForkJoinPool.java:2000) at java.base/java.util.concurrent.ForkJoinPool.helpJoin(ForkJoinPool.java:2034) at java.base/java.util.concurrent.ForkJoinTask.awaitDone(ForkJoinTask.java:423) at java.base/java.util.concurrent.ForkJoinTask.join(ForkJoinTask.java:651)...

I changed `./gradlew build` to `./gradlew build -x test` and it didn't fail, so I guess that worked.

The profile is a graph annotated with statistics on some nodes, the metrics are mostly global counters. What does the unification look like?

What data structure is the result of the unification of a graph and a collection of counters? I am serious here. I don't know what this should look like. Since...

OK, I'm beginning to understand. I think this idea has a fundamental impedance mismatch, because these profiles are per-worker but metrics are per-process (not even per-circuit although we only currently...

Here is another variation of the same problem. Without `arbitrary_precision`, both tests will pass. With `arbitrary_precision`, the integer test will pass but the real test will fail: ```rust #![allow(dead_code)] use...

I found a workaround. As a diff from the test program I posted above: ```patch diff -u /home/blp/tmp/rusttmp/src/main.rs.broken /home/blp/tmp/rusttmp/src/main.rs --- /home/blp/tmp/rusttmp/src/main.rs.broken 2025-08-05 10:18:01.282269642 -0700 +++ /home/blp/tmp/rusttmp/src/main.rs 2025-08-05 10:17:04.363135084 -0700 @@...

If you use the direct URL to the pipeline, then it doesn't time out. I think the pipeline manager should just be more patient with the timeout. On Wed, Jul...

> @blp How did you reproduce this? I've been doing a lot of profiles on slow pipelines for a week or two. I usually had to run it multiple times...

> It would be nice if we don't explicitly have to wrap things in a Runtime::init_bla closure. e.g., we can take some inspiration on how tokio does it with a...