AIXCL v1.1.68
Release v1.1.68 -- Security hardening release: removes a hardcoded Vault auth token and its dead HCL configs, narrows the Vault listener to loopback, closes a plaintext-password exposure in pgAdmin's config generator and in postgres statement logging, and aligns postgres/open-webui restart policy with the rest of the stack.
What's New in v1.1.68
Fixed
- ✅ Remove hardcoded Vault auth token and dead agent HCL configs: Three Vault Agent HCL configs carried a hardcoded auth token and were never actually invoked by any container's
command:-- every real credential path already uses thevaultCLI or the HTTP API directly. Removed the dead configs along with their dangling bind mounts and stale cross-references inCONTEXT.md/ADR docs (#1995). - ✅ Narrow Vault listener to 127.0.0.1, matching its own comment: The Vault listener bound to
0.0.0.0:8200despite a comment claiming loopback-only. Verified everyVAULT_ADDRconsumer in the repo already connects via127.0.0.1, so narrowing the bind changes no behavior while closing the gap between the stated and actual configuration (#1996). - ✅ Fail closed instead of writing a world-readable DB password: pgAdmin's server-config generator fell back to
chmod 644(world-readable) wheneverpodman unshare chownfailed, exposing a live PostgreSQL password on the host filesystem. It now fails closed, leaving the file at600host-owned and requiring the connection be added manually in the UI (#1997). - ✅ Stop postgres from logging a plaintext password via log_statement=all: The bootstrap entrypoint's password-sync
ALTER USER ... PASSWORDstatement was being logged verbatim -- password included -- into container log files whose rotation lifecycle is decoupled from the credential's own rotation in Vault. Removedlog_statement=all/log_min_duration_statement=0from both the base compose file and the postgres-ssl overlay, the latter caught by a repo-wide sweep for sibling instances of the same pattern (#2003). - ✅ Use unless-stopped restart policy for postgres and open-webui: These two services were the only ones of 17 restart-managed services set to
restart: always, meaning they would come back up after a host reboot even if the operator had intentionally stopped the stack. Aligned both withunless-stoppedto match every other service (#2004).
Documentation
Full Changelog: v1.1.67...v1.1.68