joshuaauerbachwatson / remotebuildaction Goto Github PK

The current error handling in the action-loop builder actions (written in Python) is flawed in at least these ways.

1. Many paths through the code (those involving errors detected by nim or unhandled exceptions in nim or in Python code), were coming back as timeouts. This is discussed in nimbella/nimbella-cli#163.
2. Even if the previous were corrected, many "normal error" returns (typically, due to developer errors in the submitted action) would still come back as if they were serious system errors and required system-level explorations, such as examining the activation record.

I proposed two fixes in the CLI issue, one very modest one in the issue description, and a more extensive one in a later comment. The first was tried out by Nimbella and fixed the timeout problem in cases that were of immediate interest but it turns out it was not complete enough. The second proposal was nearly complete enough but continued to refer the developer to the activation record for details that ought to have been fed back in a more intuitive way as outputs of the deployment operation.

The task now is to fix the remaining details.

The current error handling in the action-loop builder actions (written in Python) is flawed in at least these ways.

1. Many paths through the code (those involving errors detected by nim or unhandled exceptions in nim or in Python code), were coming back as timeouts. This is discussed in nimbella/nimbella-cli#163.
2. Even if the previous were corrected, many "normal error" returns (typically, due to developer errors in the submitted action) would still come back as if they were serious system errors and required system-level explorations, such as examining the activation record.

I proposed two fixes in the CLI issue, one very modest one in the issue description, and a more extensive one in a later comment. The first was tried out by Nimbella and fixed the timeout problem in cases that were of immediate interest but it turns out it was not complete enough. The second proposal was nearly complete enough but continued to refer the developer to the activation record for details that ought to have been fed back in a more intuitive way as outputs of the deployment operation.

The task now is to fix the remaining details.