fix(zenoh): reopen pooled sessions after fork

[jiajiayao]

Contribution path

• Linked issue or discussion: Closes --daemon + Zenoh: Coordinator RPC service unreachable after daemonize (Dimos.connect / out-of-process clients time out) #3395

Problem

With the Zenoh transport, dimos run ... --daemon builds the coordinator before double-forking. That can populate the process-default Zenoh session pool in the parent. The daemon grandchild inherits the Python session objects, but not Zenoh's runtime threads, so CoordinatorRPC.serve() reuses a dead session and out-of-process clients cannot reach Coordinator/ping.

The inherited pool lock is also unsafe to reuse if another parent thread held it when the process forked.

Solution

Make ZenohSessionPool process-aware:

• keep the owning PID, session cache, and lock in one process-owned state object
• replace that entire state when os.getpid() changes
• open a fresh session for the child instead of reusing or closing inherited Zenoh handles
• retain same-process session sharing and existing close_all() behavior

A regression test simulates the PID transition and verifies that the child opens a second session while leaving the inherited parent handle untouched.

How to Test

pytest dimos/protocol/service/test_zenohservice.py \
  dimos/protocol/rpc/test_spec.py \
  dimos/core/test_zenoh_transport.py \
  dimos/core/coordination/test_coordinator_rpc.py \
  dimos/core/test_daemon.py -q

Local result on macOS: 78 passed, 9 skipped. The skipped cases are existing macOS-specific RPC skips.

Also run:

ruff format --check dimos/protocol/service/zenohservice.py dimos/protocol/service/test_zenohservice.py
ruff check dimos/protocol/service/zenohservice.py dimos/protocol/service/test_zenohservice.py
mypy dimos/protocol/service/zenohservice.py
python -m compileall -q dimos/protocol/service/zenohservice.py dimos/protocol/service/test_zenohservice.py
git diff --check

All passed.

The repository-wide parallel fast suite was attempted locally but was not a clean signal on macOS: unrelated worker/relay/control E2E tests stalled or failed under the 16-worker run, and a serial test_e2e_daemon.py run stalled during worker bootstrap before executing a test. The focused Zenoh/RPC/daemon tests above were rerun after that attempt and remained green. A Linux reproduction of the actual --daemon command remains appropriate in CI or maintainer validation.

AI assistance

OpenAI Codex with GPT-5 assisted with issue analysis, implementation, tests, and PR preparation. The process-lifecycle behavior, two-file diff, and reported validation results were reviewed before submission.

Checklist

• I have read and approved the CLA.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

@@            Coverage Diff             @@
##             main    #3457      +/-   ##
==========================================
+ Coverage   74.05%   74.06%   +0.01%     
==========================================
  Files        1283     1283              
  Lines      124704   124729      +25     
  Branches    11141    11142       +1     
==========================================
+ Hits        92349    92383      +34     
+ Misses      29493    29487       -6     
+ Partials     2862     2859       -3     

Flag	Coverage Δ
OS-ubuntu-24.04-arm	70.44% <100.00%> (+<0.01%)	⬆️
OS-ubuntu-latest	72.22% <100.00%> (+<0.01%)	⬆️
Py-3.10	72.22% <100.00%> (+<0.01%)	⬆️
Py-3.11	72.22% <100.00%> (+<0.01%)	⬆️
Py-3.12	72.22% <100.00%> (+<0.01%)	⬆️
Py-3.13	72.22% <100.00%> (+<0.01%)	⬆️
Py-3.14	72.22% <100.00%> (+<0.01%)	⬆️
Py-3.14t	?

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
dimos/protocol/service/test_zenohservice.py	100.00% <100.00%> (ø)
dimos/protocol/service/zenohservice.py	84.12% <100.00%> (+9.79%)	⬆️

... and 1 file with indirect coverage changes

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.