Antifailure
GitHubDocs
Sign inStart free
Solutions · Marketplaces

Queues, workers, dual-writes, matching logic staging never reproduces.

The twin includes workers and queues. Production webhooks are blocked.

Impatient retries and multi-tab checkout become deterministic scenarios.

Matching logic that depends on queue order will not show up if staging skips workers.

Start the quickstartRequest hosted access
Trace · matching.worker
3 hops · 1 denied
  1. 01
    listings.created
    web → queue
  2. 02
    matching.worker
    queue → worker
  3. 03
    api.partners.test
    worker → partner · denied
Timing is the bug

Services, queues, and workers are a dimension staging drops.

  • Queues in the twin. Simulated streams so dual-writes and retries are visible.

  • Webhook containment. Production partner webhooks are blocked and written to the attempted-effect ledger.

  • Retry personas. Impatient users and API clients become deterministic scenarios.

Clone-local queue
t0 → now
listings.created
t+00:02
orders.paid
t+00:07
match.attempt
t+00:11
notify.worker
t+00:14
partner.webhook
blocked
00:0000:06now00:18
3 pass1 warn1 block
Restore both sides

Buyers, sellers, listings, and in-flight orders as a referential subset.

  • Both sides of the market. Buyers and sellers restored together, so a match has something to match against.

  • Run the workers. Matching, notify, and settle against clone-local queues.

  • Contain partners. Each partner host carries its own mode in the manifest, so the ones the twin simulates and the ones it refuses are written down rather than assumed.

Referential subset
2 joined · 1 miss
Buyers
match
Sellers
  • 01
    buyer_18c
    in-flightjoined
  • 02
    buyer_44a
    activejoined
  • 03
    buyer_09f
    long-tailmiss
miss
  • seller_north
    listing openjoined
    01
  • seller_helix
    listing openjoined
    02
  • order_992
    join validmiss
    03
buyer_44a ↔ seller_helix1 miss
Compare

Duplicate events, missed matches, and irreversible writes in the oracle.

  • Rolling deploys. Old and new schema coexistence is exactly what a disposable twin is for.

  • Duplicate events. Visible when workers actually run.

  • Compare. The oracle diffs the twin's writes against the baseline run.

Oracle · dual-write
old worker / new worker

Baseline

Candidate

queued
listings.created
queued
listings.created
match.ok
orders.paid
match.miss
orders.paid
sent
notify.worker
skipped
notify.worker
Branched database
one row the oracle returns
miss
rowbasecand
order_992okmiss
Keep reading
Next

Know what happens before you deploy.

Create a disposable production twin for every risky change. Catch migration failures before they reach customers.

Start the quickstartRequest hosted access