mirror of
https://github.com/Ayymoss/Plutainer
synced 2026-08-24 14:32:06 -04:00
d117dede7d
Six pieces of work that turned out to share a lot of ground. T7x no longer needs Xvfb. Passing -headless makes it skip Sys_CreateConsole and print to stdout instead of building a Win32 console window, which is the only reason a display was ever required. Verified against a clean volume; the images now ship no X server at all, saving ~340MB. Seed configs are vendored under seed-configs/ instead of being wget'd from six third-party repos at build time. Any one of those disappearing broke the build for every game, and alterware/t7x 404'd while this was being written. The RUN string never changed either, so BuildKit cached the layer indefinitely and upstream edits stayed invisible. tools/refresh-seeds.sh pins each repo to a commit, records it in seed-configs/<game>/SOURCE, and blanks the placeholder credentials upstream ships (an RCON password everyone knows, and in cod4x's case a g_password that locked players out). The healthcheck no longer needs an RCON password. It asks the server for its status and requires a loaded map, trying getstatus then getinfo: IW5 and T5 only answer the latter, while T7x's getinfo reports the lobby's map rather than the running one. Every bundled config ships rcon_password empty, so the old RCON-based check could never pass on a first run. pyquake3 is trimmed to the two paths actually used, and now tolerates the stray byte T7x prefixes to its statusResponse - which is why RCON never worked on T7x at all. T5 and T6 gate unauthenticated getinfo on the RCON whitelist, so a sidecar IW4MAdmin completes the RCON handshake and then fails on unanswered getinfo. The sender is the container's Docker gateway, assigned at runtime, so it cannot be written into a config file: resolve_rcon_whitelist_args detects it and passes +rconWhitelistAdd. PLUTAINER_RCON_PASSWORD writes a password into the config, opt-in and never destructive. CoD4x is new, and is the one family that does not use Wine - upstream ships a native Linux server, and the Windows build dies at window creation exactly as T7x did. That makes it amd64-only, since the binary is 32-bit x86. Two assets come from the client release because the server release has no fastfiles. Game logs are rotated at 64MB. A large enough log stops CoD4x logging permanently: the writer hands each chunk to fwrite, and on failure prints a warning and drops the data with the buffer already advanced. Rotation is copy-truncate, safe because every engine opens its log with fopen(path, "ab"). A marker line is written after truncating, and into any empty log, because IW4MAdmin treats a zero offset as "no position yet" and re-syncs instead of reading - which silently swallowed the first batch of events after each rotation, and the first map's events on every fresh deployment. Also fixes the arm64 healthcheck start period, which was 1m against amd64's 5m and would have marked a first-run IW4x download unhealthy mid-download, and bumps Hangover 11.4 -> 11.9. |
||
|---|---|---|
| .. | ||
| refresh-seeds.sh | ||