Summary

Follow-up to #1592. Reduces the on-disk logging footprint from 8 transports per logger version (16 total, counting v1 and v2) down to 3 rotated files, eliminates unbounded files, adds retention limits, and auto-cleans legacy files from hosts upgrading across CADT versions.

Context for why there were so many files:

Inspecting a real deployment's ~/.chia/mainnet/cadt/v1/logs/:

Winston 3's per-transport level is independent of the logger's level, so debug-%DATE%.log with level: 'debug' captured verbose/debug regardless of APP.LOG_LEVEL.

Changes

New transport set (per logger version)

transports: [
  new DailyRotateFile({
    filename: `${logDir}/application-%DATE%.log`,
    datePattern: 'YYYY-MM-DD',
    level: 'debug',                         // captures error..debug stream
    zippedArchive: true,
    maxSize: '20m',
    maxFiles: '30d',                        // NEW: bounded retention
    utc: true,
    format: format.combine(format.json()),
  }),
],
exceptionHandlers: [
  new DailyRotateFile({ /* exceptions-%DATE%.log, 20m, 30d, zipped */ }),
],
rejectionHandlers: [
  new DailyRotateFile({ /* rejections-%DATE%.log, 20m, 30d, zipped */ }),
],

Disk usage is now bounded at ~600 MB per logger version (30 × 20 MB compressed) instead of unbounded.

Dropped transports

• error.log — subset of application-%DATE%.log; filter via jq '.level=="error"' or journalctl -p err
• combined.log — pure duplicate
• debug-%DATE%.log — merged into application-%DATE%.log at debug level
• exceptions.log and rejections.log — unbounded duplicates of the rotated siblings

Startup cleanup of legacy files

New cleanupObsoleteLogFiles(logDir) runs at logger creation and removes files superseded by rotated equivalents:

• error.log, combined.log, exceptions.log, rejections.log (the four unbounded files)
• debug-%DATE%.log, debug-%DATE%.log.N, debug-%DATE%.log.gz (matched via /^debug-.*\.log(\.\d+)?(\.gz)?\$/)

Behavior guarantees:

• Best-effort: missing files, missing directories, and per-file unlink errors never throw; they're collected and logged as a warning via the new logger once it's constructed.
• Idempotent: second run on a clean directory removes 0 files.
• Narrow: files whose names loosely resemble debug files but don't match the regex (debug-backup.tar, debug.txt, other.log) are preserved.

Other adjacent concerns that this PR deliberately does NOT address (to keep diff narrow):

• Making maxFiles configurable via APP.LOG_RETENTION_DAYS — fine with 30 as a default; can add a knob later if operators ask
• Removing `console.trace(error)` calls from other controllers — separate cleanup
• Updating docker/k8s volume sizing recommendations — followup docs PR if desired

Verification

Ran two dedicated smoke tests against winston@3.19.0 + winston-daily-rotate-file@5.0.0.

Cleanup logic (seeded with the user's actual production filenames plus adversarial non-log filenames):

== removed (9) ==
combined.log
debug-2026-04-13.log
debug-2026-04-16.log
debug-2026-04-16.log.1
debug-2026-04-16.log.gz
debug-2026-04-20.log
error.log
exceptions.log
rejections.log

== kept ==
application-2026-04-13.log, application-2026-04-20.log,
exceptions-2026-04-20.log, rejections-2026-04-20.log,
debug-backup.tar, debug.txt, other.log

second-run removed: 0 errors: 0
missing-dir removed: 0 errors: 0

New application transport with logger level info, transport level debug, all 6 levels emitted:

=== application-2026-04-21.log ===
{"level":"error",  "message":"TEST_ERROR",   ...}
{"level":"warn",   "message":"TEST_WARN",    ...}
{"level":"info",   "message":"TEST_INFO",    ...}
{"level":"verbose","message":"TEST_VERBOSE", ...}
{"level":"debug",  "message":"TEST_DEBUG",   ...}

silly is correctly excluded (transport level is debug; if operators want silly, they can set APP.LOG_LEVEL=silly and it will reach the journalctl Console transport from #1592 without cluttering disk).

Impact on existing deployments

• Upgrading hosts lose 5 files on first start (the four unbounded files + any existing debug-*.log* files). All content in them is either duplicated elsewhere on disk or available in journalctl.
• Disk usage drops. In the investigated deployment, the unbounded files had already reached 46 MB combined and would continue growing; the rotated debug files had accumulated 8 days × up to 3 files/day × up to 20 MB.
• Operational commands change slightly: grep targets that hit error.log or combined.log need to move to application-*.log or to journalctl.
• Nothing in this PR changes what Winston emits to journalctl — fix(API): capture staging commit errors in logs and stream all logs to journalctl #1592 owns that stream.

Test plan

• npm run test passes locally
• Start CADT against an existing ~/.chia/mainnet/cadt/v1/logs/ that contains the old files; verify:
  • The five legacy file classes are removed on startup
  • New log lines land in application-%DATE%.log at all levels error..debug
  • A structured info entry reporting the cleanup count and filenames is present
• Start CADT against a fresh / empty log directory; verify no startup warnings about missing files
• Simulate a read-only log directory (or make one file immutable with chattr +i); verify the logger still starts and emits a structured warning listing the files it couldn't remove
• Confirm maxFiles: '30d' behavior by advancing the system clock or manually creating older dated files and verifying they're eventually purged