From f85b497794a9c0d2b5a666d2b68c340dc9b32964 Mon Sep 17 00:00:00 2001 From: Agents Agent Date: Sun, 16 Aug 2026 21:13:32 +0200 Subject: [PATCH] fix(flaresolverr): schedule on the always-on amd64 node and raise the CPU cap The Deployment pinned only to personal-stack/site: enschede, so the pod landed on enschede-pi-1. Solving a Cloudflare Turnstile challenge runs headless Chrome, which the arm64 Pis cannot finish inside Prowlarr's 60 s proxy timeout: challenge-solving requests took 70-85 s and returned 500 "Error solving the challenge. Timeout after 60.0 seconds.", surfacing in Prowlarr as "Unable to connect to proxy: Http request timed out". /health kept answering 200 in milliseconds throughout, so the container stayed Ready and the failure never appeared as a restart or a probe error. The pod now pins to enschede-t1000-1. It stays in Enschede because a residential IP draws far fewer challenges than a datacenter address; the other amd64 Enschede nodes are desktops that power off when not in use, so the always-on t1000 is the only stable amd64 target. The CPU limit goes from 1000m to 3000m. Chrome saturated the single-core cap during a solve, accumulating ~1100 s of cgroup throttling. The request stays at 500m because the process idles between indexer queries. --- .../apps/stateless/flaresolverr/deployment.yaml | 16 +++++++++++++++- 1 file changed, 15 insertions(+), 1 deletion(-) diff --git a/platform/cluster/flux/apps/stateless/flaresolverr/deployment.yaml b/platform/cluster/flux/apps/stateless/flaresolverr/deployment.yaml index 540cf4c7..b0acd8f3 100644 --- a/platform/cluster/flux/apps/stateless/flaresolverr/deployment.yaml +++ b/platform/cluster/flux/apps/stateless/flaresolverr/deployment.yaml @@ -14,7 +14,16 @@ spec: app.kubernetes.io/name: flaresolverr spec: nodeSelector: + # Enschede keeps the solver on a residential IP, which draws far + # fewer Cloudflare challenges than a datacenter address would. personal-stack/site: enschede + # Solving a Turnstile challenge runs headless Chrome, which the + # arm64 Pis cannot finish inside Prowlarr's 60 s proxy timeout — + # scheduled on enschede-pi-1 every challenge timed out and + # Prowlarr reported the indexer proxy as unreachable. The other + # amd64 Enschede nodes are desktops that power off when not in + # use, so pin to the always-on t1000 server. + personal-stack/node: enschede-t1000-1 containers: - name: flaresolverr # Upstream resumed active development (v3.5.0, 2026-05; now @@ -46,7 +55,12 @@ spec: cpu: 500m memory: 768Mi limits: - cpu: 1000m + # Chrome saturated the old 1-core cap while solving a + # challenge (~1100 s of cgroup throttling on the Pi), which + # is what pushed solve times past the client timeout. The + # request stays low because the process idles between + # indexer queries; only the burst needs the headroom. + cpu: 3000m memory: 2Gi --- apiVersion: v1