Run controlled workloads that demonstrate Forge's scheduling, dependency, retry, and recovery behavior.
Start with Parallel Processing to watch the DAG, then try Failure & Retry Recovery to inspect attempt history. The full showcase combines both with an intentionally failed branch.
This is a shared live demo: runs are visible to every visitor. Up to three workflows run at once, with a short pause between launches and a daily compute budget.
Inspect a 30-second Tears of Steel excerpt, transcode three segments in parallel with FFmpeg, then merge a playable MP4 with audio.
Expected result: All five tasks succeed and the finished video appears on the workflow page.
The Docker image includes a 30-second 1280×572 Tears of Steel excerpt with audio. Allow up to two minutes.
A 23-task release pipeline: merge three data streams, fan out a test matrix, recover an unstable integration, and publish to two destinations while an experimental branch exhausts its retries.
Expected result: The main release and audit succeed. The experimental check fails after three attempts and its two descendants are skipped, so the overall workflow intentionally ends FAILED.
Uses the shared Docker volume for retry recovery across workers. Allow 1–3 minutes.
Run Forge's Java, C++ and Python verification workload as a distributed DAG.
Expected result: Succeeds when all three verification branches pass.
Docker workers include the source and build tools. Runs Java unit tests and C++ tests plus Python syntax checks; allow up to ten minutes.
Inject a first-attempt failure and watch Forge recover through a durable retry.
Expected result: Succeeds after the unstable task recovers on its second attempt.
Uses the shared Docker volume so a retry can recover on another worker.
One preparation step releases three independent branches, then a final gate joins their results.
Expected result: All tasks succeed; publish waits for every processing branch.
Python 3 on workers. Parallelism depends on available capacity.
Three independent inputs finish at different times. Watch the join stay blocked until the slowest is ready.
Expected result: The join succeeds only after all three inputs succeed.
Python 3 on workers.
A failing quality check prevents a shared release gate from running while an independent branch finishes normally.
Expected result: Intentionally fails: release is skipped, but the independent branch succeeds.
Python 3 on workers.
Fail the first stage of a four-stage pipeline and follow the skip through every downstream dependency.
Expected result: Intentionally fails: the remaining three tasks are skipped without execution.
Python 3 on workers.
A consistently failing task uses all three attempts before Forge skips its dependent publish step.
Expected result: Intentionally fails after three attempts; publish never runs.
Python 3 on workers. Allow time for retry backoff.
Release eight bounded jobs together and watch available worker slots fill, then join at a final completion gate.
Expected result: All tasks succeed. Jobs queue when available slots are fewer than eight.
Python 3 on workers. Open Workers during the run to watch load.