1. Intent
Dagster
Job: rnaseq_job
Asset: star_align
Run ID: #4521
Executor: k8s_executor

Resources (requested):
CPUs: 64
Memory: 488 GB
2. Execution (Dagster Run / Infrastructure)
Execution Reality
Dagster run: rnaseq_job / run #4521
Step: star_align
Executor: k8s_executor
Pod: star-align-4521
Instance / node: r5.16xlarge
Container image: ../biocontainers/star
Duration: 47 min
Dagster executed the step using statically defined resource requests. Runtime behavior inside the container was not observable from Dagster events or logs alone.
3. Reality
Observed inside Dagster step star_align
Dagster job rnaseq_job → run #4521 → step star_align
CPU
8%
Memory
18%
Disk I/O
114%
Diagnosis: Step is I/O-bound. Increasing CPU or memory in the Dagster run config will not improve runtime.
4. The Fix
Recommended Dagster resource override for step star_align:
Compute resource: io_optimized_compute
Requested CPUs: 8
Requested memory: 64 GB
Storage: Local NVMe scratch
Node / instance: r6id.2xlarge
Scope: Step-level resource override (not global job config)
This change updates the Dagster step's compute resource. Same job, same asset — ~85% lower compute cost and 3.7× faster runtime.