zzstoatzz.io
atproto relay implementation in zig
zlay.waow.tech
docs: document memory tuning (thread stacks + c_allocator)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
+10
-2
docs/deployment.md
+10
-2
docs/deployment.md
···
71
71
- **monitoring**: prometheus + grafana via kube-prometheus-stack
72
72
- **terraform**: `infra/zlay/` in the relay repo
73
73
74
+
## memory tuning
75
+
76
+
two changes brought steady-state memory from ~6.6 GiB down to ~2.9 GiB at 2,738 connected hosts:
77
+
78
+
**thread stack sizes.** zig's default thread stack is 16 MB. with ~2,750 subscriber threads that maps 44 GB of virtual memory. most threads just read websockets and decode CBOR — 2 MB is generous. all `Thread.spawn` calls now pass `.{ .stack_size = 2 * 1024 * 1024 }`. the constant is defined in `main.zig` as `default_stack_size` for the threads spawned there; other modules use the literal directly.
79
+
80
+
**c_allocator instead of GeneralPurposeAllocator.** GPA is a debug allocator — it tracks per-allocation metadata and never returns freed small allocations to the OS. since zlay links glibc (`build.zig:42`), `std.heap.c_allocator` gives us glibc malloc with per-thread arenas, madvise-based page return, and production-grade fragmentation mitigation.
81
+
74
82
## resource usage
75
83
76
84
| metric | value |
77
85
|--------|-------|
78
-
| memory | ~1.8 GiB steady state (1486 subscribers) |
86
+
| memory | ~2.9 GiB steady state (~2,750 hosts) |
79
87
| CPU | ~1.5 cores peak |
80
88
| limits | 8 GiB memory, 250m CPU request |
81
89
| PVC | 20 GiB (events + RocksDB collection index) |
82
-
| postgres | ~131 MiB |
90
+
| postgres | ~238 MiB |
83
91
84
92
## git push
85
93