fabi/Mangalord
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(vision-manager): memory-pressure yield gate (+ retain analysis gate) #8

Merged
fabi merged 2 commits from feat/vision-mem-yield into main 2026-06-16 14:34:25 +00:00
Conversation 0 Commits 2 Files Changed 5 +342 -9
4 changed files with 342 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Showing only changes of commit 0a9e727530 - Show all commits

12
.env.example
View File

@@ -218,3 +218,15 @@ VISION_MANAGER_DATABASE_URL=
# VISION_START_HEALTH_TIMEOUT Max seconds to wait for /health 200 after start. Default 300 (cold model load).
# VISION_RESPECT_CRAWL_MUTEX 1 = don't start vision while a crawl runs (RAM mutex on the 8 GiB box). Default 1.
# VISION_MAX_UPTIME >0 = force-stop vision after N seconds running (backstop). Default 0 (disabled).
#
# Memory-pressure yield — the dynamic generalization of the crawl mutex. The
# manager reads host-wide used% (from /proc/meminfo MemAvailable) and stops a
# RUNNING vision when memory gets tight, so a spike elsewhere (a crawl, CI, a
# backend burst) can finish instead of the kernel OOM-killer shooting something
# stateful. Two watermarks give hysteresis; the cooldown prevents stop/restart
# thrash when vision itself is the hog. See VISION-MEMORY-YIELD.md.
# VISION_MEM_YIELD_ENABLED 1 = enable the gate, 0 = disable entirely. Default 1.
# VISION_MEM_HIGH_WATERMARK_PCT used% >= this → actively stop a running vision. Default 92.
# VISION_MEM_LOW_WATERMARK_PCT used% >= this → inhibit starts (keep HIGH - LOW >= ~10 so the band beats jitter). Default 80.
# VISION_MEM_YIELD_COOLDOWN Seconds after a pressure-stop during which restart is refused regardless of backlog. Default 300.
# VISION_MEM_POLL_INTERVAL Mem sub-poll cadence, seconds (<= VISION_POLL_INTERVAL); catches spikes between backlog ticks. Default 5.

230
VISION-MEMORY-YIELD.md Normal file
View File

@@ -0,0 +1,230 @@
# Vision-manager — memory-pressure yield (design outline)
**Status:** implemented in `vision-manager/manager.sh` (feat/vision-mem-yield).
Verifications V1–V4 (§5) all passed against the current backend — expired
leases are reclaimed continuously on every claim poll (not boot-only), connection
errors requeue with backoff, the readiness gate parks before leasing, and admin
re-enqueue recovers any straggler — so no backend changes were required.
`RESPECT_CRAWL_MUTEX` is **kept** for now (retiring it is deferred until the
watermarks are tuned on the real box).
**Extends:** `VISION-AUTOSCALE.md` (the `vision-manager` sidecar) and
`vision-manager/manager.sh`.
**One-liner:** let the manager *proactively stop* `mangalord-vision` when host
memory gets dangerously tight, so a transient spike elsewhere (a Chromium crawl,
a CI build, a backend burst) can complete instead of the kernel OOM-killer
shooting something stateful (postgres / backend).
---
## 1. Motivation
`llama-server` has no idle-unload and pins ~4.4 GiB while up. On the 8 GiB Pi
the existing autoscaler already stops it when the **backlog** is empty and (new)
when **analysis is disabled**. Neither reacts to *actual memory pressure*:
- `mem_limit: 6g` on the vision service caps only **vision's own** growth. It
does nothing about **aggregate host** pressure across all services.
- The static `RESPECT_CRAWL_MUTEX` only **defers a start** while a crawl runs —
it can't stop a vision that's **already running**, and it's blind to every
non-crawl memory consumer (CI build, backend spike, a second crawl worker).
- The only backstop under true pressure today is the **kernel OOM killer**,
which may pick postgres or the backend (long-lived, stateful, painful) rather
than the one process that is cheap and safe to lose.
A manager-driven stop is the friendly version: it picks the **right,
restartable victim** and does it **gracefully (SIGTERM)** before the kernel does
something violent.
This is the **dynamic generalization of `RESPECT_CRAWL_MUTEX`**: "back off when
RAM is actually tight" strictly dominates "never co-run with a crawl." If this
ships, `RESPECT_CRAWL_MUTEX` can likely be retired.
---
## 2. Feasibility (confirmed)
`/proc/meminfo` inside the manager container reports **host-wide** memory (it is
not namespaced), so the manager already sees the truth with **zero new
privilege** — no socket-proxy `docker info`, no host mount. Verified live from
inside `vision-manager`:
```
MemTotal: 8256576 kB (7.87 GiB)
MemAvailable: 1754384 kB → ~79% used, 21% available (vision up + 560 queued)
```
**Use `MemAvailable`, never `MemFree`.** Linux spends "free" RAM on page cache by
design, so a `MemFree`-based "used %" sits near 90 % during normal operation and
would false-trigger constantly. `MemAvailable` already nets out reclaimable
cache. The honest metric is:
```
used_pct = 100 * (1 - MemAvailable / MemTotal)
```
Read it in bash with `awk` over `/proc/meminfo` (both values are in kB):
```sh
read_mem_used_pct() {
awk '/^MemTotal:/{t=$2} /^MemAvailable:/{a=$2}
END { if (t>0) printf "%d", 100*(1-a/t); else print "ERR" }' /proc/meminfo
}
```
---
## 3. Behaviour
A third gate in the existing poll loop, evaluated **before** the
backlog/enable start/stop logic:
1. **Stop gate (active).** If `used_pct >= HIGH_WATERMARK` (e.g. 92) **and**
vision is running → `docker stop mangalord-vision`, record a cooldown
deadline, log loudly. This is the new capability the crawl mutex never had.
2. **Start inhibit.** While `used_pct >= LOW_WATERMARK` (e.g. 80) **or** a
cooldown is in effect → refuse to start vision even if backlog > 0 (same
shape as the crawl-mutex deferral: log once per episode, not every tick).
3. **Resume.** Once `used_pct < LOW_WATERMARK` **and** the cooldown has elapsed,
normal autoscaler behaviour resumes (start on backlog, idle-stop, etc.).
**Hysteresis is mandatory** (two distinct watermarks). A single threshold
oscillates around the boundary.
**The cooldown is mandatory** and is the subtle part. If vision *itself* is the
4.4 GiB hog, stopping it drops pressure → manager sees backlog → restarts it →
pressure climbs → stop … infinite thrash, each cycle paying the ~minutes cold
reload. The cooldown (refuse-to-restart for N minutes after a pressure-stop,
*regardless of backlog*) is what gives the **other** workload — the one that
actually needed the RAM — room to finish before vision muscles back in. The
yield only makes sense when something else needs the memory; the cooldown
encodes that.
### Gate ordering in the loop
```
loop every POLL_INTERVAL:
used = read_mem_used_pct()
pending = pending_analysis()
if not analysis_enabled(): pending = 0 # existing gate
if mem_yield_active(used): pending = 0 # NEW: inhibit starts
if used >= HIGH_WATERMARK and vision running: # NEW: active stop
stop_vision(); start_cooldown(); continue
... existing start-on-backlog / idle-debounce-stop logic on `pending` ...
```
Note the symmetry with the analysis-disabled gate already in `manager.sh`:
forcing `pending = 0` reuses the existing idle/stop path for the *inhibit* case,
while the explicit `stop_vision()` handles the *active* case the idle path can't.
---
## 4. Proposed configuration (all env, matching manager.sh style)
| Var | Default | Meaning |
|---|---|---|
| `MEM_YIELD_ENABLED` | `1` | Master switch for the whole feature. |
| `MEM_HIGH_WATERMARK_PCT` | `92` | `used_pct ≥` this → active-stop a running vision. |
| `MEM_LOW_WATERMARK_PCT` | `80` | `used_pct ≥` this → inhibit starts (no restart until below). |
| `MEM_YIELD_COOLDOWN` | `300` | Seconds after a pressure-stop during which restart is refused regardless of backlog. |
| `MEM_POLL_INTERVAL` | `POLL_INTERVAL` | Optional faster cadence for the mem check (pressure can spike between 20 s ticks). |
Watermarks want tuning on the real box; 92/80 are starting points. Keep
`HIGH − LOW ≥ ~10` so the band is wider than normal jitter.
---
## 5. Required verifications (do these BEFORE building)
The queue is built for expendability — `crawler_jobs` is a lease queue:
```
state | attempts | max_attempts(=5) | leased_until | last_error | scheduled_at
```
A graceful `docker stop` (SIGTERM → clean llama-server exit) means at most the
**one** in-flight `analyze_page` job is affected; the readiness gate
(`ANALYSIS_VISION_HEALTH_URL`) then parks the worker so it won't hammer the dead
endpoint or burn the other queued jobs. But two things must be confirmed in
`backend/src/analysis/daemon.rs` (not visible from the DB schema alone):
- [ ] **V1 — reclaim cadence.** The `bugfix/crawler-recovery-hardening` branch
added *reclaim-orphaned at startup*. Confirm expired leases are **also**
reclaimed on the normal claim poll (i.e. the claim query includes something
like `OR (state='running' AND leased_until < now())`). If reclaim is
**boot-only**, a pressure-kill strands that one job as `running` with a dead
lease until the next backend restart → the page silently never analyses.
**This decides how aggressive we can safely be.**
- [ ] **V2 — error path requeue.** When the worker's HTTP call to vision is
refused (vision killed), confirm the job is returned to `pending`
(`attempts++`) and not marked permanently `failed` on a connection error.
A connection-refused/transport error should be *retryable*, distinct from a
model-returned bad response.
- [ ] **V3 — park does not deadlock.** Confirm that when the readiness gate
parks the worker, it is **not holding an un-renewed lease** that blocks the
job indefinitely (it should either release on park or let the lease expire +
be reclaimed per V1).
- [ ] **V4 — attempts accounting.** A pressure-kill is "our fault," yet still
burns one of the 5 attempts. Confirm repeated yields on the *same* in-flight
page can't quietly exhaust retries → permanent `failed`. Mitigations if it's a
concern: bump `max_attempts`, or rely on the admin re-enqueue (all/manga/
chapter) you already have to recover stragglers.
**Empirical check** (once V1–V3 look right): with analysis on and one job
`running`, `docker stop mangalord-vision`, then watch
`SELECT state, attempts, last_error FROM crawler_jobs WHERE state IN
('running','pending') AND payload->>'kind'='analyze_page'` — the killed job
should reappear as `pending` with `attempts` incremented, and resume when vision
restarts. This is the acceptance test for the whole feature.
---
## 6. Pitfalls
- **`MemFree` vs `MemAvailable`** — using `MemFree` makes it fire constantly
(page cache). Already addressed by §2, but it's the #1 way to get this wrong.
- **Thrash / feedback loop** — if vision is the hog, stop→restart oscillates.
The cooldown (§3) is the guard; without it the feature is worse than nothing.
- **Spikes between ticks** — a Chromium burst can blow past 92 % inside a 20 s
poll gap and the kernel OOM-kills before the manager's next read. The valve
*reduces* OOM risk, it doesn't *eliminate* it. Consider a shorter
`MEM_POLL_INTERVAL` for just the mem read, and treat this as defence-in-depth
alongside (not instead of) `mem_limit` and swap/zram headroom.
- **Measuring the wrong scope** — confirm `/proc/meminfo` in the container keeps
showing **host** totals under this kernel/cgroup setup (it does today:
MemTotal 8256576 kB == the Pi). If a future runtime namespaces it, the metric
silently becomes the container's tiny cgroup and the gate never/always fires.
- **Cold-start cost** — every pressure-stop pays ~minutes to mmap + warm 3.2 GiB
on the next start. Don't set the watermarks so tight that normal operation
trips them; the backlog will stall during each reload.
- **Interaction with the idle-stop and disable gate** — make sure the three
reasons vision can be down (idle-drained, analysis-disabled, memory-yield) are
**distinctly logged** (flip-only, like the existing gates) so a stop is never
misdiagnosed. A single muddy "stopped vision" line will cost debugging time.
- **Double-stop / start races** — `docker stop` then an immediate backlog-driven
`docker start` in the same or next tick. The cooldown plus checking
`vision_running` before acting prevents this; don't drop those guards.
- **OOM-killer still owns the tail** — `mem_limit: 6g` caps vision's own growth
but the host can still be pushed over by the *sum* of everything else. This
feature lowers the probability of a bad OOM kill; pairing it with a little
**swap or zram** gives the kernel a cushion to survive the gap between a spike
and the manager's reaction.
---
## 7. Recommendation
Worth building, as a **third gate** in the existing `manager.sh` poll loop:
`MemAvailable`-based metric, two watermarks (hysteresis), a restart cooldown,
an **active stop** (the capability the crawl mutex lacks), and distinct
flip-only logging. It supersedes `RESPECT_CRAWL_MUTEX` (which can then be
removed).
**Gate it on verification V1 first** — whether expired leases are reclaimed
continuously or only at boot decides whether a yielded job cleanly retries or
strands. If V1/V2 hold, this is a low-risk, high-value safety valve; if reclaim
is boot-only, fix that in the backend before enabling memory-yield, otherwise
each pressure-stop quietly drops a page until the next backend restart.
Keep the manager dumb: this is ~30 lines of bash (one `awk` reader, two
watermark comparisons, one cooldown timestamp) — no new dependencies, no new
privilege, no web surface.

8
docker-compose.yml
View File

@@ -168,6 +168,14 @@ services:
START_HEALTH_TIMEOUT: ${VISION_START_HEALTH_TIMEOUT:-300}
RESPECT_CRAWL_MUTEX: ${VISION_RESPECT_CRAWL_MUTEX:-1}
MAX_UPTIME: ${VISION_MAX_UPTIME:-0}
# Memory-pressure yield — stop vision when the HOST is short on RAM so a
# spike elsewhere can finish without the kernel OOM-killer. See
# VISION-MEMORY-YIELD.md.
MEM_YIELD_ENABLED: ${VISION_MEM_YIELD_ENABLED:-1}
MEM_HIGH_WATERMARK_PCT: ${VISION_MEM_HIGH_WATERMARK_PCT:-92}
MEM_LOW_WATERMARK_PCT: ${VISION_MEM_LOW_WATERMARK_PCT:-80}
MEM_YIELD_COOLDOWN: ${VISION_MEM_YIELD_COOLDOWN:-300}
MEM_POLL_INTERVAL: ${VISION_MEM_POLL_INTERVAL:-5}
networks:
- default # reach postgres + the vision container by name
- vision-internal # reach the socket-proxy

93
vision-manager/manager.sh
View File

@@ -30,6 +30,17 @@ MAX_UPTIME="${MAX_UPTIME:-0}" # >0: force-stop after N idle-or-not
# together OOM the 8 GiB box). Set to 0 on roomier hosts.
RESPECT_CRAWL_MUTEX="${RESPECT_CRAWL_MUTEX:-1}"
# ---- Memory-pressure yield (see VISION-MEMORY-YIELD.md) --------------------
# Dynamic generalization of the crawl mutex: stop a RUNNING vision when the
# HOST is actually short on RAM, so a transient spike elsewhere (a crawl, CI,
# a backend burst) can finish instead of the kernel OOM-killer shooting
# something stateful. Uses host-wide MemAvailable from /proc/meminfo.
MEM_YIELD_ENABLED="${MEM_YIELD_ENABLED:-1}" # 0 disables the whole gate
MEM_HIGH_WATERMARK_PCT="${MEM_HIGH_WATERMARK_PCT:-92}" # used% >= → active-stop a running vision
MEM_LOW_WATERMARK_PCT="${MEM_LOW_WATERMARK_PCT:-80}" # used% >= → inhibit starts (hysteresis)
MEM_YIELD_COOLDOWN="${MEM_YIELD_COOLDOWN:-300}" # s after a pressure-stop before restart allowed
MEM_POLL_INTERVAL="${MEM_POLL_INTERVAL:-5}" # mem sub-poll cadence, <= POLL_INTERVAL
PSQL=(psql "$DATABASE_URL" -At -v ON_ERROR_STOP=1)
log() { echo "[vision-manager] $(date -u +%FT%TZ) $*"; }
@@ -53,6 +64,18 @@ crawl_running() {
AND state = 'running';" 2>/dev/null || echo "ERR"
}
# ---- Memory pressure -------------------------------------------------------
# Host-wide used% from MemAvailable (NOT MemFree, which sits near "full" thanks
# to page cache and would false-trigger constantly). /proc/meminfo in this
# container reports HOST totals (it is not namespaced under this kernel).
# Emits the "ERR" sentinel on a malformed read so callers can ignore it.
read_mem_used_pct() {
# Require BOTH fields: a partial read, or a pre-3.14 kernel without
# MemAvailable, would otherwise leave `a` at 0 → used=100% → a false stop.
awk '/^MemTotal:/{t=$2} /^MemAvailable:/{a=$2}
END { if (t>0 && a!="") printf "%d", 100*(1-a/t); else print "ERR" }' /proc/meminfo
}
# ---- Docker helpers --------------------------------------------------------
# `docker inspect .State.Running` → true/false; empty if the container is
# absent (defined elsewhere and not yet created) — treated as not-running.
@@ -84,7 +107,8 @@ start_vision() {
}
stop_vision() {
log "stopping $VISION_CONTAINER (idle ${STOP_DEBOUNCE}s)"
local reason="${1:-idle ${STOP_DEBOUNCE}s}"
log "stopping $VISION_CONTAINER ($reason)"
if docker stop "$VISION_CONTAINER" >/dev/null 2>&1; then
return 0
fi
@@ -92,12 +116,43 @@ stop_vision() {
return 1
}
# ---- Memory-yield gates ----------------------------------------------------
# Active stop: when the host is over the HIGH watermark and vision is running,
# SIGTERM it (the friendly, right-victim alternative to an OOM kill) and arm a
# cooldown. Cheap on the hot path — only inspects docker once used% is already
# over the mark. Reads `mem_yield_until`/`idle_for`/`up_for` from the loop scope.
mem_check_and_stop() {
[ "$MEM_YIELD_ENABLED" = "0" ] && return 0
local used; used="$(read_mem_used_pct)"
case "$used" in ''|*[!0-9]*) return 0 ;; esac # ignore a malformed read
if [ "$used" -ge "$MEM_HIGH_WATERMARK_PCT" ] && [ "$(vision_running)" = "true" ]; then
if stop_vision "memory-yield: used ${used}% >= ${MEM_HIGH_WATERMARK_PCT}% high watermark"; then
mem_yield_until=$(( $(date +%s) + MEM_YIELD_COOLDOWN ))
idle_for=0
up_for=0
fi
fi
}
# Start inhibit: echo 1 while a cooldown is in effect OR used% is at/above the
# LOW watermark. The two distinct watermarks give the mandatory hysteresis; the
# cooldown gives whatever needed the RAM room to finish before vision restarts.
mem_yield_active() {
[ "$MEM_YIELD_ENABLED" = "0" ] && { echo 0; return; }
[ "$(date +%s)" -lt "$mem_yield_until" ] && { echo 1; return; }
local used; used="$(read_mem_used_pct)"
case "$used" in ''|*[!0-9]*) echo 0; return ;; esac
if [ "$used" -ge "$MEM_LOW_WATERMARK_PCT" ]; then echo 1; else echo 0; fi
}
# ---- Main loop -------------------------------------------------------------
log "started: container=$VISION_CONTAINER health=$VISION_HEALTH_URL poll=${POLL_INTERVAL}s debounce=${STOP_DEBOUNCE}s docker_host=$DOCKER_HOST"
log "started: container=$VISION_CONTAINER health=$VISION_HEALTH_URL poll=${POLL_INTERVAL}s debounce=${STOP_DEBOUNCE}s docker_host=$DOCKER_HOST mem_yield=$([ "$MEM_YIELD_ENABLED" = 0 ] && echo off || echo "${MEM_LOW_WATERMARK_PCT}/${MEM_HIGH_WATERMARK_PCT}% cooldown=${MEM_YIELD_COOLDOWN}s mempoll=${MEM_POLL_INTERVAL}s")"
idle_for=0 # seconds the queue has been empty while vision is running
up_for=0 # seconds vision has been running (for MAX_UPTIME backstop)
crawl_deferred=0 # 1 while a start is held off by the crawl mutex (log on flip only)
mem_yield_until=0 # epoch deadline of the active memory-yield cooldown (0 = none)
mem_inhibited=0 # 1 while starts are inhibited by memory pressure (log on flip only)
while true; do
pending="$(pending_analysis)"
@@ -114,6 +169,22 @@ while true; do
;;
esac
# Memory-yield gates (before the backlog logic). The active stop handles a
# running vision the idle path can't; the inhibit forces pending=0 so the
# existing idle-debounce path keeps it down (same shape as the crawl mutex).
mem_check_and_stop
running="$(vision_running)" # refresh: the active stop may have just flipped it
if [ "$(mem_yield_active)" = "1" ]; then
if [ "$mem_inhibited" != "1" ]; then
log "memory-yield: inhibiting vision start (used >= ${MEM_LOW_WATERMARK_PCT}% or cooldown active)"
mem_inhibited=1
fi
pending=0
elif [ "$mem_inhibited" = "1" ]; then
log "memory-yield: pressure cleared — start inhibit lifted"
mem_inhibited=0
fi
if [ "$running" = "true" ]; then
up_for=$((up_for + POLL_INTERVAL))
else
@@ -160,12 +231,24 @@ while true; do
# Backstop: bound how long vision can stay up regardless of debounce state.
if [ "$MAX_UPTIME" -gt 0 ] && [ "$running" = "true" ] && [ "$up_for" -ge "$MAX_UPTIME" ]; then
log "MAX_UPTIME ${MAX_UPTIME}s reached; force-stopping $VISION_CONTAINER"
if stop_vision; then
if stop_vision "MAX_UPTIME ${MAX_UPTIME}s reached"; then
idle_for=0
up_for=0
fi
fi
sleep "$POLL_INTERVAL"
# Sleep until the next backlog poll, but re-check memory pressure on the
# faster MEM_POLL_INTERVAL sub-cadence so a spike between backlog ticks can
# trigger an active stop before the kernel OOM-killer does. Degenerates to a
# single sleep when MEM_POLL_INTERVAL >= POLL_INTERVAL.
slept=0
while [ "$slept" -lt "$POLL_INTERVAL" ]; do
step="$MEM_POLL_INTERVAL"
remaining=$((POLL_INTERVAL - slept))
[ "$step" -gt "$remaining" ] && step="$remaining"
[ "$step" -lt 1 ] && step=1
sleep "$step"
slept=$((slept + step))
mem_check_and_stop
done
done