Goal
Validate needle.server on the two target platforms that issue #2 did not cover: a systemd Linux machine and a Raspberry Pi 5.
Scope
- Run
setup.sh, the test suite, and the systemd service on Linux x86_64 and on Raspberry Pi 5 (linux-arm64).
- Measure the engine thread count for one runner, and find out if the runner flag
--threads changes it. On a 32-thread Linux machine the shared library started 12 threads for each engine, and 8 concurrent engines overloaded the machine. Do not run more than one engine at a time on a shared machine until this is known.
- Set the concurrency limit for each platform from measurements. The macOS value is 2.
- Watch worker memory. The reported
peak_ram_mb increased by approximately 1 MB for each worker during 100 requests. If it continues, recycle a worker after a fixed number of requests.
Done when
The service answers a correct tool call on each of the two platforms, and PLAN.md records the measured thread count, the concurrency limit, and the memory result.
Goal
Validate needle.server on the two target platforms that issue #2 did not cover: a systemd Linux machine and a Raspberry Pi 5.
Scope
setup.sh, the test suite, and the systemd service on Linux x86_64 and on Raspberry Pi 5 (linux-arm64).--threadschanges it. On a 32-thread Linux machine the shared library started 12 threads for each engine, and 8 concurrent engines overloaded the machine. Do not run more than one engine at a time on a shared machine until this is known.peak_ram_mbincreased by approximately 1 MB for each worker during 100 requests. If it continues, recycle a worker after a fixed number of requests.Done when
The service answers a correct tool call on each of the two platforms, and
PLAN.mdrecords the measured thread count, the concurrency limit, and the memory result.