fix: scope pool totals to --dax-path and serialize RpcClient requests

[seohui-XCENA]

Summary

Fixes two issues surfaced while wiring the LMCache MP L1 fullness gauges: get_stats/get_usage summed CXL capacity across every resource-manager pool (foreign DAX devices inflate the eviction-watermark total), and the shared ZMQ REQ socket in RpcClient corrupts when two threads issue RPCs concurrently (eviction loop + metrics scrape both calling get_stats).

Key Changes

maru_server — scope _pool_totals to the server's --dax-path pools

• request_alloc only claims regions from the configured pools, but the totals summed every RM pool. On a host with a foreign device this inflates cxl_pool.free_size, which clients use as the eviction-watermark total.
• Observed on a 3-device host: a 3.5 TiB /dev/dax0.0 pushed the watermark total to ~4 TiB against a 466 GiB usable pair (dax1.0+dax2.0) — usage ratio stuck at ~5%, so client-side LRU eviction would never engage and stores would run into allocation failure instead.
• _pool_totals() now filters by the server's dax_path scope; unrestricted servers keep the all-pools behavior. Covers both get_stats.cxl_pool (watermark input) and get_usage.pool_total/pool_free (marutop).

maru_handler — serialize RpcClient requests with a per-call lock

• ZMQ sockets are not thread-safe and REQ enforces strict send→recv alternation; overlapping exchanges degrade both calls to spurious timeouts plus socket resets.
• connect/close/_send_request (including the reset paths) now run under one per-instance lock; concurrent callers queue. _reset_socket is only called with the lock held.
• No measurable overhead: A/B against a local REP echo server shows sequential mean 59.3µs (main) vs 59.1µs (this branch); an uncontended acquire+release is ~113ns (~0.2% of a roundtrip). Under 2-thread contention, main fails 30/100 calls while this branch fails 0/100 — the lock only ever blocks a caller in the case where main returned garbage.
• The handler's stats flusher uses its own dedicated RpcClient (separate lock), so it never contends with the control path. AsyncRpcClient is untouched.

Test Plan

• Unit tests added/updated
  • test_pool_totals_scoped_to_configured_dax_paths: foreign pool excluded from get_usage and get_stats.cxl_pool
  • TestRpcClientThreadSafety::test_concurrent_send_requests_are_serialized: 8 threads, no overlapping send→recv exchange
• Existing tests pass (pytest -v) — all consumer suites green (test_maru_server, test_rpc_client, test_server, test_usage_monitor, test_rpc_handler_mixin, test_allocation_manager, test_rpc_async_client, test_shm_client, test_shm_get_dax_path; 217 tests). test_maru_handler.py/test_thread_safety.py bus-error on the dev host identically on clean main (live-RM + DAX mmap environment issue, unrelated to this change).
• E2E: against the live resource manager on the 3-device host, a --dax-path dax1.0,dax2.0 server now reports cxl_pool 466 GiB vs 4042 GiB unrestricted.

Related Issues

Related: #60 (scopes the cxl_pool totals introduced there)

[github-actions]

Coverage Report

File	Stmts	Miss	Cover	Missing
__init__.py	6	0	100%
__main__.py	3	3	0%	5, 7–8
allocation_manager.py	129	11	91%	30–31, 44–45, 52, 237–238, 268–271
client.py	187	8	96%	88, 130, 147–148, 311–313, 319
config.py	46	1	98%	72
constants.py	9	0	100%
device_scanner.py	94	31	67%	25, 100, 102–104, 106, 114–123, 125, 127, 129, 134–143, 145, 147
handler.py	524	79	85%	125, 140, 151–158, 166–168, 176, 187–192, 197, 222, 249–250, 254, 294, 298–299, 304, 326, 333–334, 338–344, 347, 351–353, 369, 371–372, 429, 450, 460–461, 567–569, 576, 698–701, 704, 715–718, 724, 1056–1060, 1066, 1077–1081, 1087, 1166, 1171, 1213
ipc.py	275	2	99%	365, 441
kv_manager.py	108	0	100%
logging_setup.py	19	0	100%
protocol.py	233	0	100%
resource_manager_installer.py	103	13	87%	80–86, 167, 169–172, 187
rpc_async_client.py	190	0	100%
rpc_async_server.py	111	0	100%
rpc_client.py	71	0	100%
rpc_client_base.py	106	10	91%	185, 221–222, 233–234, 306–307, 311–312, 375
rpc_handler_mixin.py	106	19	82%	154–156, 159–161, 219–222, 227, 231–235, 256–257, 263
rpc_server.py	64	0	100%
serializer.py	81	0	100%
server.py	170	13	92%	64–65, 73, 168, 172, 243, 247, 265–266, 357–359, 444
stats_manager.py	95	0	100%
types.py	60	1	98%	145
uds_helpers.py	13	0	100%
memory
__init__.py	5	0	100%
allocator.py	55	0	100%
mapper.py	128	2	98%	229, 296
owned_region_manager.py	101	1	99%	212
types.py	62	0	100%
TOTAL	3174	194	94%

Tests	Skipped	Failures	Errors	Time
690	4 💤	0 ❌	0 🔥	7.161s ⏱️