日志记录#

日志订阅者通过 Python 的标准 logging 模块为存储、检索、查找、L1 和 StorageManager 事件发出调试级别的消息。

当安装 OpenTelemetry 时，init_logger 会自动附加一个 OTel LoggingHandler，以便将日志记录转发到任何配置的 OTel LoggerProvider。该处理程序遵循 LMCACHE_LOG_LEVEL 环境变量。

LMCACHE_LOG_LEVEL=DEBUG lmcache server ...

关键日志消息：

级别	消息
INFO	`Stored N tokens in X seconds`
INFO	`Retrieved N tokens in X seconds`
INFO	`Prefetch request completed (L1+L2): N/M prefix hits`
DEBUG	`MP store start: session=... device=...`
DEBUG	`MP retrieve end: session=... retrieved_count=...`

Extra Logging (periodic transfer stats)#

Opt-in periodic INFO-level stats, aimed at watching a running server from its logs without a metrics stack:

lmcache server ... --enable-extra-logging --extra-logging-interval 10

Every interval, the server logs — per GPU — the L0<->L1 store/retrieve activity of the last window and the cumulative totals since start (operation count, tokens, size in GB, and average copy throughput in GB/s):

L0<->L1 stats [cuda:0] last 10.0s: store ops=12 tokens=24576 size=1.50GB avg_copy=12.34GB/s | retrieve ops=3 tokens=6144 size=0.40GB avg_copy=18.10GB/s
L0<->L1 stats [cuda:0] cumulative: store ops=120 tokens=245760 size=15.02GB avg_copy=12.10GB/s | retrieve ops=31 tokens=63488 size=4.11GB avg_copy=17.92GB/s

The eviction loop also reports L1 memory usage at the same interval:

L1 memory usage: 45.20/100.00 GiB (45.2%)

Notes:

Transfer-stat lines are only emitted for windows with activity; the L1 memory usage line is emitted every interval regardless of load.
avg_copy measures throughput of the GPU-stream copy time only (not wall clock), and shows n/a when no timed operation completed in the window.
--extra-logging-interval values below 1.0 are limited by the 1 Hz internal heartbeat.
--enable-extra-logging works even with --disable-logging, but conflicts with --disable-observability (the server raises a ValueError at startup).