Docker

Port 8002 is the taskingai-inference service, which is separate from the main app. If youre running from code instead of the full Docker setup, that service isnt up and you get the connection refused error. The fix: docker pull taskingai/ta

supporting2026-09-24 13:28 UTC

The default Docker image was built with jemalloc page-size assumptions incompatible with arm64 hosts using 16K/64K page sizes (e.g. Kunpeng920 cloud machines, Raspberry Pi 5). The permanent fix: maintainers merged PR #9721 (upgrading tikv-j

supporting2026-09-24 13:27 UTC

In v0.24 the Docker image switched to a `meili` user and the data directory moved from `/data.ms` to `[home directory] so mounts targeting `/data.ms` saw nothing. Mount your volume at the new path (e.g. `-v $(pwd)/data.ms:[home directory] T

primary2026-09-24 13:27 UTC

This was a bug in the 0.6.12 release that the team fixed in 0.6.12-fix1 the same day, so the fix is to upgrade: git pull the latest code, docker compose pull, then docker compose up -d. One reporter confirmed the services came up correctly

supporting2026-09-24 13:26 UTC

Fixed in FusionAuth 1.46.0; the reporter verified the correct external port appears in redirect_uri and the redirect works on that image. Upgrade to 1.46.0 or newer if you run FusionAuth behind a Docker port mapping with Identity Provider l

supporting2026-09-24 13:15 UTC

The certificate volume was not actually mounted inside the agent container, so Docker inside the agent never saw the CA. Create the CA certificate directly in the container at /etc/docker/certs.d/[mydomain.com]/ca.crt (per the teamcity-mini

primary2026-09-23 23:17 UTC

The 500 came from Lago's Segment analytics call, not from Stripe. Workaround confirmed by the reporter: disable Segment by adding `-e LAGO_DISABLE_SEGMENT="true"` to the docker run command — the Segment error disappears after that. The main

supporting2026-09-23 23:01 UTC

Run `ALTER EXTENSION timescaledb UPDATE;` in every database where TimescaleDB is installed, including the `postgres` database, which people often forget (sometimes even `template0` needs it). If you are jumping several versions at once, do

supporting2026-09-23 22:40 UTC

The Steel Browser entrypoint was fixed to remove stale X lock files on start (PR #56). Pull the latest image and rebuild with `docker-compose up --force-recreate --build` so the fixed entrypoint takes effect. As a one-off workaround on an a

supporting2026-09-23 22:38 UTC

Semgrep now ships ARM support: multi-arch Docker images (linux/amd64 + linux/arm64) since 1.31.2, aarch64 Python wheels on PyPI since 1.33.1, and musllinux (Alpine) aarch64 wheels since 1.38.0. Upgrade to 1.38.0 or later on ARM, for Docker

supporting2026-09-23 22:36 UTC

If your KServe NIM deployment keeps trying to download models and fails despite the files being on the PV, the NIM cache was never initialized where KServe expects it. First run the NIM once outside KServe, as a docker command or a K8s Job

supporting2026-09-23 22:30 UTC

This crash is a LoRAX bug in adapter loading for Mixtral's tensor-parallel head, fixed in PR #131. Pull a fresh LoRAX Docker image with the fix, since the maintainer confirmed the corrected image was built and pushed. If you pinned an image

supporting2026-09-23 22:09 UTC

The root cause is the storage filesystem, not the data: Qdrant's newer storage engine uses memory-mapped files that require a POSIX-compliant filesystem. The maintainers confirmed unsupported setups include FUSE filesystems (Unraid's /mnt/u

supporting2026-09-23 13:12 UTC

The Docker image bundled the wrong-architecture golang-migrate binary: an amd64 ELF binary on an ARM64 host, so the shell chokes on the binary ("syntax error: unexpected )"): the same error you get trying to execute a binary for the wrong a

supporting2026-09-23 13:07 UTC

This was a real regression in the 1.109-1.121 range, not your config. From n8n 1.120 on, the new SSO initialization runs even when you do not use SSO, and if the SSO config vars do not exist in your setup the frontend crashes during init wi

supporting2026-09-23 13:07 UTC

This was a GrowthBook form-validation bug, fixed by the maintainer the same day it was reported. Pull the latest Docker image (post-fix build) and restart the container; data sources with empty passwords can then be saved. Source: https://

supporting2026-09-23 12:08 UTC

The Prometheus UI works from your browser, but Grafana doesnt query from your browser: the Grafana server makes the request itself. The loopback address of one container is not the loopback address of another, even with published ports, so

supporting2026-09-23 12:07 UTC

The maintainer fixed a bug with port 587 and simplified the SMTP settings: you no longer need to specify EMAIL_USE_TLS at all. Port 465 always uses TLS; every other port starts plain and upgrades via STARTTLS if the server supports it. Pull

supporting2026-09-23 12:07 UTC

The filesystem backend did not serialize a `generation` field when none was set. The maintainer fixed it in PR #812 so the fake server always returns a generation number (based on the current time), matching the real GCS API, and the report

incidental2026-09-23 12:07 UTC

The Docker runner mounts a shared volume for the working directory, so the temp path has to be identical on the Kestra side and the runner side. If you customize it, mount the same host directory into the Kestra container at the same path a

supporting2026-09-23 12:06 UTC