Skip to content

Troubleshooting

Under normal operation, InaCAWO is designed as an automated, self-correcting forecast system with built-in backup options in each job's decision tree. This page covers common failure modes and how to diagnose them using ROME logs and Zabbix status messages.

Log analysis procedure

When ROME's monitor job (M1) flags an error, follow this procedure:

  1. Check verify directory — inspect $RT_LOGSNOW/verify/ for flagged log file names
  2. Open the cycle log — the actual log is in $RT_LOGSNOW (e.g. $RT_LOGS/cycles/2024052812/)
  3. Trace to cron log — identify the cron job schedule from $RT_SCRIPTS/cron/cawofcst_prod@mdclogin0_utc.tab, then review $RT_LOGSCRON/
  4. Follow the job tree — top-level worker scripts may reference lower-level logs; follow references down the tree

Example: a missing hycom_download.success file means HYCOM pre-processing (Job R2) will not run. Trace back to Job D3's cron log in $RT_LOGSCRON/ and the ingest log in $RT_LOGS/ingest/.

Typical errors

Global ocean dataset problems

Affects MERCATOR (Job D4/R1) and HYCOM (Job D3/R2) downloads and pre-processing.

Cause Symptom Impact
Provider API change Download script failures Scripts revised when MERCATOR/CMEMS protocols update
Network interruption Partial downloads, retry loops Jobs retry until timeout; may switch to backup source
Bad data (NaNs) MATLAB pre-processing issues Filtered by nearest-neighbor fill; rare downstream ROMS errors

The system is resilient — losing one day of HYCOM files typically does not disrupt service because Mercator is prioritized. Zabbix workflow files track which global ocean source is active per cycle.

ECMWF-IFS timeliness (00/12 UTC only)

ECMWF is the preferred atmospheric source for on-cycle runs. If ECMWF files are not available within the wait window, the system falls back to GFS for both ICs and BCs.

Scenario Behavior
ECMWF 24 h files arrive on time ECMWF used for entire 10-day forecast
ECMWF delayed beyond wait period GFS fallback activated; flag recorded in Zabbix
ECMWF arrives in irregular batches Pre-processing backlog may delay model chunks; forecast marked "slow"

The forecast runs in six chunks (0–24, 24–48, 48–96, 96–144, 144–192, 192–240 h). Each chunk waits up to 30 minutes for ECMWF-based files before switching to GFS.

ROMS model blow-ups

ROMS is sensitive to CFL violations when ocean velocities exceed the stability limit (transport distance > grid spacing in one timestep).

Cause Example
Pre-processing memory contention on login node ROMS CFL error on off-cycle run (documented June 2023 SAT)
Bad boundary condition data Silent MATLAB pre-processing failure propagating NaNs
Missing spun-up initial conditions Using uninterpolated global model ICs

When a ROMS blow-up occurs, the affected cycle's forecast may fail. Subsequent cycles typically recover if the root cause (e.g. login node contention) is resolved.

HPC interruptions

Issue Effect Recovery
Compute node failure Single cycle may be missed Next cycle uses previous restart files
Mellanox Infiniband issues Batch job failures Resubmit; ROME monitor flags the error
Disk space low ROME cleaning alerts (Job M3) Automatic purge per mdclogin0_disks.cfg thresholds

Zabbix status monitoring

Zabbix consumes text status files produced by tracking jobs (D5–D10) and forecast scripts. Key status categories:

Category What it tracks
Disk space warnings Available space below configured thresholds
Pre-processing / analysis Six-hourly analysis cycle completion
10-day forecast runs On-cycle (00/12 UTC) timeliness and fallback flags
Off-cycle forecast runs 90-hour (06/18 UTC) completion
Post-processing FTP transfer and plot generation status

Failure file

InaCAWO maintains a failure file that records cycle-level failures. Operators should review this file alongside ROME verify logs when investigating missed forecast cycles.

When to intervene manually

Manual intervention is rarely needed. The system is designed to:

  • Switch between primary and backup global data sources automatically
  • Continue forecast production even when one ocean source is unavailable
  • Alert operators via ROME monitor and Zabbix without stopping the pipeline

Intervention may be warranted when:

  • Multiple consecutive cycles fail (possible HPC outage)
  • Crontab changes are being deployed (install incrementally)
  • Persistent ECMWF delivery issues degrade forecast quality (contact data provider)
  • Disk space cleaning thresholds need adjustment

Info

For operational support, contact produksi.maritim@bmkg.go.id.

References

  • OM_InaCAWO_Related-Software-Manual.pdf — Sections 3.2–3.3 (SOPs, error handling, Zabbix)
  • 20230615_InaCAWO_SAT_Operation_Test_M1_623_v1.0.pdf — operational acceptance test results