Error Handling

Error sensitivity

By default, R steps fail only on errors (non-zero exit code). The error_on field lets you treat warnings or messages as failures too:

- id: check
  check: "."
  error_on: warning    # treat R warnings as failures

Level	Fails on
`error` (default)	R errors (non-zero exit)
`warning`	R errors + warnings
`message`	R errors + warnings + messages

This is implemented via withCallingHandlers() wrapping the generated R code. For script: steps, a thin wrapper sources the user script inside the handler.

R version enforcement

Constrain the R version at DAG level:

r_version: ">=4.1.0"

By default, a version mismatch warns but continues. To fail hard:

r_version: ">=4.1.0"
r_version_strict: true

The check runs at DAG start via Rscript --version, before any step executes.

Retries

Steps can retry on failure with configurable backoff:

- id: query
  script: etl/query.R
  retry:
    limit: 3               # retry up to 3 times
    backoff: exponential    # 1s, 2s, 4s, 8s...
    max_delay: 60s          # cap delay at 60 seconds

Field	Default	Description
`limit`	0	Number of retries (total attempts = limit + 1)
`backoff`	`linear`	`linear` or `exponential`
`max_delay`	none	Cap on delay between retries

R error extraction

When an R step fails, daggle extracts the actual R error message from stderr and shows it in daggle status:

STEP     STATUS  ERROR
extract  failed  Error in readRDS("missing.rds"): cannot open connection

No need to dig through log files for the error message.

Timeouts

Each step can specify a timeout. On expiry:

SIGTERM sent to the entire process group
5-second grace period
SIGKILL if still running

- id: long-query
  script: etl/query.R
  timeout: 10m

No orphaned R processes – process group killing ensures child processes are cleaned up too.