Skip to content
pgflow

⚠️ Project Status

Edge Worker should be considered an advanced Proof of Concept.

It is not production-ready at this time.

I am actively working on resolving several known issues.

PostgresError: DbHandler exited

When processing a high volume of jobs with increased concurrency (10 or more), the system occasionally fails with the following error:

[Error] Timeout-triggered archive failed: PostgresError: DbHandler exited
    at ErrorResponse (https://deno.land/x/[email protected]/src/connection.js:791:26)
    at handle (https://deno.land/x/[email protected]/src/connection.js:477:6)
    at data (https://deno.land/x/[email protected]/src/connection.js:318:9)
    at https://deno.land/x/[email protected]/polyfills.js:138:30
    at Array.forEach (<anonymous>)
    at call (https://deno.land/x/[email protected]/polyfills.js:138:16)
    at success (https://deno.land/x/[email protected]/polyfills.js:98:9)
    at eventLoopTick (ext:core/01_core.js:168:7) {
  name: "PostgresError",
  severity_local: "FATAL",
  severity: "FATAL",
  code: "XX000"
}

This issue appears to be related to abruptly terminated SQL connections in Session Mode. When this error occurs, it prevents the system from spawning new instances.

Postgres Deadlocks

In high-concurrency scenarios, I’ve observed occasional deadlocks. These occur due to race conditions between message archiving and message pickup when visibility timeouts expire (educated guess).

The planned solution involves implementing worker-side retries for SQL queries.

Planned Architecture Improvements

Following the resolution of current issues, a major architectural refactor is planned. The main goals are to:

  • Implement proper dependency injection
  • Introduce a factory/builder pattern
  • Enable easy component swapping, including:
    • MessageExecutor (required for pgflow orchestrator integration)
    • Polling mechanism (replacing ReadWithPollPoller with ListenNotifyPoller for improved performance)