Run monitor continuously by default (--interval optional, 0 = back-to-back)

[berendt]

Closes #27

What changed

Monitor mode (neutron monitor) previously required --interval and rejected a zero or non-positive value. This branch makes --interval optional with a default of 0, which now means continuous: each iteration starts the moment the previous one finishes. Paced runs (--interval > 0) are unchanged.

The change lands in three commits:

275de60 — Run monitor continuously by default (--interval 0 = back-to-back)

• cmd/openstack-tester/monitor.go: drop MarkFlagRequired("interval") so the flag is optional; its default was already 0. The flag usage string now documents 0 = continuous.
• monitorConfig.validate() now rejects only a negative interval (< 0) instead of a non-positive one (<= 0); the error message changes from "must be positive" to "must not be negative". A zero interval is valid and means back-to-back.
• Startup log: renders interval=continuous when the interval is 0, so the line stays self-explanatory; the value is otherwise cfg.interval.String().
• runMonitorLoop: adds a minFailureBackoff (5s) floor on the post-failure wait. In continuous mode (interval and error-wait both 0), a fast-failing dependency — e.g. an unreachable Keystone that refuses auth in ~1ms — would otherwise hot-loop, pinning a CPU and flooding logs; the floor makes it back off. --error-wait remains the explicit brake.
• Command/doc-comment wording updated to say "continuously or on a cadence".

018c967 — Default testbed-monitor to a continuous MONITOR_INTERVAL of 0

• Makefile: MONITOR_INTERVAL default flips from 5m to 0, so make testbed-monitor runs iterations back-to-back and feeds the local OTEL stack a steady stream of data. MONITOR_INTERVAL=5m restores paced behaviour. The cadence echo annotates both zeros (interval 0 (0 = continuous)) so it stays readable.

7cffd7d — Document the continuous monitor default in README and dashboards

• README.md: reword the command synopsis (--interval now optional), the cadence semantics, and the testbed-monitor walkthrough to lead with the new default; note that --error-wait is the brake in continuous mode.
• contrib/otel/dashboards/*.json: the three sampling caveats no longer claim a 5m default, since metrics now flow steadily under the continuous default and only move per-iteration on a paced run.

Tests

cmd/openstack-tester/monitor_test.go gains coverage for the new semantics (continuous pacing, validate() accepting 0 and rejecting negatives, and the failure-backoff floor), all driven on the injected fake clock — no cloud required.

Divergence from the issue

• The issue asked only to make --interval optional with 0 = back-to-back. The implementation additionally introduces a fixed 5s minFailureBackoff floor for failed iterations, to prevent a hot-loop when both --interval and --error-wait are 0. This is a safety guard implied by "continuous by default" rather than a literal ask.
• Note for reviewers: commit 275de60's message says "the loop and clock need no changes", but that same commit does add the minFailureBackoff floor to runMonitorLoop (documented in the function's doc comment). The message narrates the primary mechanism; the guard is the one loop change that shipped alongside it.

---

Implemented by planwerk-agent 6b3f3ea with Claude:claude-opus-4-8