SNOW-2070718: Should Blocked be considered an error, as it is still running?

[Jmaharman]

I've just seen this code and thought it odd that Blocked is considered an error, when it is actually still running? What I mean is that the Blocked status could change, which means it should be in the IsStillRunning?

internal static bool IsStillRunning(QueryStatus status)
{
    switch (status)
    {
        case QueryStatus.Running:
        case QueryStatus.ResumingWarehouse:
        case QueryStatus.Queued:
        case QueryStatus.QueuedReparingWarehouse:
        case QueryStatus.NoData:
            return true;
        default:
            return false;
    }
}

internal static bool IsAnError(QueryStatus status)
{
    switch (status)
    {
        case QueryStatus.Aborting:
        case QueryStatus.FailedWithError:
        case QueryStatus.Aborted:
        case QueryStatus.FailedWithIncident:
        case QueryStatus.Disconnected:
        case QueryStatus.Blocked:
            return true;
        default:
            return false;
    }
}

[sfc-gh-dszmolka]

hello - thank you for creating this issue with us and drawing attention to this one. Looks like it's indeed unexpected.
Looked at the JDBC driver implementation and the Python driver implementation and indeed looks like we consider BLOCKED (= 'The state when a statement is waiting on a lock on resource held by another statement') as 'still running' in those connectors.

Will address here too.

[sfc-gh-dszmolka]

opened a draft PR at #1166, but it needs to be discussed internally with the team, as BLOCKED status seems to be divergently classified between the various Snowflake drivers, and we need probably have consensus on what we eventually want to do with it ;)

[sfc-gh-dszmolka]

sadly, with this change, all tests fail (and they consistently pass when BLOCKED is moved back to isError).
it is also not consistent on how we handle BLOCKED status. In some drivers, it's Error. In other drivers, it's Running. In a third group of drivers, it's again 'Error because it's not specifically kind of Running'.

So what seemed like an easy fix, is not; and probably first we would need to get consensus internally what exactly we want to do which this (and pontentially, other) status and implement it consistently across all drivers. Cannot estimate any kind of timeline on that one, but it's for sure of no high priority.