request: cancel query.

[hobord]

Add option to cancel long run query. context cancelation...

[kndndrj]

makes sense, will look into it once I have some time.

[MattiasMTS]

I plan to add this to this branch https://github.com/MattiasMTS/nvim-dbee/tree/mattias/progress_tick.

However, the ticker works as expected when the query is successful, i.e. the query function isn't exiting here https://github.com/kndndrj/nvim-dbee/blob/master/dbee/main.go#L128-L130. Because if the query is "malformed", e.g. you're querying a table that doesn't exist, then the timer will continue counting until you execute another query... :(

This is of course a behaviour we don't want. I tried to add boolean channels in the query function but that delayed the point of having go routine in the background while the frontend waits.

@kndndrj do you have any ideas on how you would implement this? I'm thinking I'd have to add this behaviour in the backend using something like:

for {
        select {
        case <-ticker.C:
          elapsed := time.Since(start)
          // Convert elapsed time to milliseconds
          elapsedTimeMillis := float64(elapsed) / float64(time.Millisecond)

          // Update the Neovim buffer with the elapsed time
          // You might need to implement this part according to your Neovim setup
          updateProgressInBuffer(elapsedTimeMillis)

          // Check if the query execution has completed
          if queryCompleted {
            logger.Debugf("%q executed successfully", method)
            err := callbacker.TriggerCallback(callbackId)
            if err != nil {
              logger.Error(err.Error())
            }
            logger.Debugf("%q successfully triggered callback", method)
            return
          }
        }
      }
    }()

    logger.Debugf("%q returned successfully", method)
    return nil
  }

and then add an option if you press <C-c> The progress bar cancels -> and the query cancels.

Perhaps adding a method that checks if the go-routine-query is still running might be worth doing, or? Polling this method with reasonable intervals would allow us to control this behaviour and we separate the control flows better.

[kndndrj]

@kndndrj do you have any ideas on how you would implement this?

Well, afaik the best way in go to cancel something is to cancel the context. So the first thing about this is to replace all the crappy context.TODOs with the actual context, which is created on query execution and passed down the chain.

This also means that pretty much all Conn methods should recieve context as a parameter.

[MattiasMTS]

@kndndrj do you have any ideas on how you would implement this?

Well, afaik the best way in go to cancel something is to cancel the context. So the first thing about this is to replace all the crappy context.TODOs with the actual context, which is created on query execution and passed down the chain.

This also means that pretty much all Conn methods should recieve context as a parameter.

Are we sure enough that cancellation of the context leads to cancellation of the Query? I think this is the best general approach to it. Any idea on how you want to craft this out? Using channels, callbacks, etc :)

[kndndrj]

If I recall correctly, the cancelation of context is interpreted by different drivers in different ways - so there is not a 100% guarantee that the query gets cancled. However I think that is also the only way of canceling a database/sql request.

as for the implementation here are my thoughts:

• context passing part shouldn't be too hard to do
• a new context gets created in Conn on every exexution and is stored until this query is the active one. then we can add a method to conn that cancles the current context
• the tricky part (not necessarly just a problem with this issue) is that there can be multiple parallel queries running on the same connection. I've been thinking about a sort of "active queries" dashboard for some time, vut didnt come up with anything concrete yet.

These are just my overall thoughts, unfortunately I can't spend the time I want on this project, so it might have to wait for a while though :/

[kndndrj]

This has been addressed in refactor pull request. Closing.