Montage: project on four GCP machines, finish locally

Montage across four GCP machines and local execution. Nodes in the same column can run in parallel once their dependencies complete.

This example preserves all 58 real Montage activities. The twelve mProject activities are split into four groups of three on separate on-demand GCP e2-medium workers; the other 46 use the local execution resource. The image ovvesley/akoflow-wf-montage:050d carries the input FITS images, headers, lookup tables, and Montage executables. AkôFlow transfers the projected FITS outputs into the local activities' workspaces. The workflow pins the published image by digest, and the bundle registers it as an artifact with an additional SIF build for Apptainer environments.

Cloud cost and validation

This plan can create four billable GCP VMs. Its predicted time is a sum of historical per-activity measurements on a different machine (c3d-standard-16); its compute cost excludes startup, disk, and network charges. Neither figure is an observed result. Check cloud quota, pricing, cleanup, and the local Docker runtime before submitting.

Download the complete bundle

Bundle README and prerequisites ↓Execution scope ↓Network topology ↓Workflow with 58 commands and named data handoffs ↓Manual GCP/local plan ↓Original Montage DAG ↓Historical activity runtimes ↓Workflow generator ↓Cloud-capacity preparation ↓Submission script ↓

The bundle README explains the instance-specific resource IDs, local Docker requirement, import order, submission command, and post-run checks.

AkôFlow Desktop

Use Desktop

Register the Montage image in Artifacts and start its SIF build. In gcp-environment, check the four e2-medium capacity targets and the online cloud connection. These are schedulable resources; no VM exists until execution needs one. Import the workflow and manual plan using the downloaded files, then inspect all 58 assignments before starting a real run. The expected final file is mosaic-color.png. In the run view, check the activity DAG, Data tab, transfer records, and cloud instance lifecycle.

API

Use the API

Follow the API access guide and the bundle README. Register the pinned image with register-artifact.sh, set the daemon's approved public IP in AKOFLOW_SSH_CIDR, then create four one-instance cloud targets with prepare-capacity.sh. Import scope.json, network.json, workflow.json, and plan.json in that order. Submit exactly one run. The example IDs refer to a particular AkôFlow instance and must be reviewed before reuse.

sh akoflow-examples/real/montage-cloud-local/register-artifact.sh
sh akoflow-examples/real/montage-cloud-local/prepare-capacity.sh
sh akoflow-examples/real/montage-cloud-local/submit-run.sh montage-58-four-vm-local-050d-run-v1

What to verify

The plan places three projections on each of four GCP machines and 46 downstream activities locally. The workflow names 138 file handoffs rather than using generic dependency.dat records. A completed run must have 58 successful activities, a non-empty mosaic-color.png, and recorded projected-FITS transfers. All four GCP VMs must be destroyed after execution. No completed observed run is claimed yet; the historical reference CSV and manual plan prediction are not a result.