v1.76.8

v1.76.8 (2026-10-20)

Release Highlights

• Collectors honor an explicit RUN_STARTUP=false and report when they skip a startup run (#1018) - Fixes #1009, Fixes #1004
  • Collectors in the all-in-one omnibus image now honor an explicit COLLECTOR_RUN_STARTUP=false and the matching variables for the optional collectors. Before this release the SMART and performance collectors ran on every container start regardless of that setting, and supplying a cron schedule made the existing guard pass anyway.
  • Behavior change for Unraid all-in-one installs. An install created from the previous Unraid template has COLLECTOR_RUN_STARTUP explicitly set to false, because that was the template default. Those installs stop collecting at startup once updated, and collect only on their cron schedule. Change that field to true to keep the previous behavior. The template now ships true for new installs. No other deployment is affected: an unset or empty variable keeps the long-standing startup run.
  • Every collector entrypoint now logs when it skips startup collection, and names the variable responsible. A container that collects nothing on start used to say nothing at all, which was hard to tell apart from a permissions failure. The standalone images also name the cron schedule they will next run on, and the omnibus image reports when a collector has no schedule and therefore will not run at all.
  • Write performance benchmarks now run without manual configuration. Automatic mount point detection could never succeed for a whole-disk device, so the collector fell back to raw device I/O where write tests are disabled. Write benchmarks previously ran only for users who had configured performance.mount_points by hand. Mount points are now resolved from the mount table, matching the device or any of its partitions.
  • A concurrent InfluxDB bucket conflict is treated as success, which removes a race during startup when several collectors create the same bucket.
  • New troubleshooting documentation for NVMe drives under device_cgroup_rules. The guide now explains why a b 259:* rwm rule grants the NVMe namespace block devices but not SMART, which the collector reads from the controller character device, and why cgroup rules grant permission without creating the device node.