x/tools/gopls: add support for multiple concurrent clients

[leitzler]

As per discussion on Slack with @ianthehat & @myitcv this is a feature request for gopls to support multiple concurrent clients.

Vim users often have multiple instances of vim running at the same time, and starting/exiting is a natural part of the workflow. Currently there is a 1 to 1 mapping between vim and gopls (at least when using govim as plugin). It would be really useful to be able to share the same cache and avoid waiting for warmup each time.

See govim/govim#491 as well as #32750 (comment) for details.

[gopherbot]

Thank you for filing a gopls issue! Please take a look at the Troubleshooting section of the gopls Wiki page, and make sure that you have provided all of the relevant information here.

[leitzler]

@gopherbot, please add label FeatureRequest

[ianthehat]

In its default state gopls supports a specific header based JSON stream of the LSP on stdin /stdout. In this mode it only supports a single client as stdin/stdout cannot be multiplexed.

It also has the flags -listen and -remote, which are designed to allow it to communicate using sockets. -listen only applies to the serve verb, and causes it to listen for incoming connections rather than stdin/stdout. In this mode it supports multiple clients. The -remote flag can be supplied to any verb, and tells it how to talk to a gopls that was started with serve -listen. If you use it with serve then you have a gopls that listens on stdin/stdout and forwards commands to the remote gopls. The intent is that you use this mode from within an editor to talk to a gopls server.

The protocol spoken between a -remote and -listen gopls is not defined, and never will be, we only support it as a way of intercommunication, not as an API surface. This is because to achieve some of its goals it will have to have significant extensions to the protocol, and may mutate some of the data on the way through. Part of the reason for this is that it should be feasible to have the server run on a separate machine, and it may not have access to the same file system, or it may know the files by different absolute paths etc. These features require a reliable way of translating paths, and also the remote file extension so the true server can ask the intermediary gopls for file contents. It may also be necessary to have some forms of caching and cancelling within the intermediary.

The current state is that we use this mode only for debugging. It only gets fixed when we need to use it to debug a problem, and even then it does not get properly fixed. It does mostly work, but there are things like straight proxying of shutdown message causes the shared server to quit when any of clients does.

There are also design issues still to fix, things like should we support some kind of "discovery" protocol, should we have a mode where we start a server if one is not running but connect to it if it is, when all the clients go away should the server shut down again, how do we manage the cache so we dont explode because of lots of clients over a very long time, how do we prevent one client from starving the others, how do we manage the config and environment of the server correctly etc

[myitcv]

Thanks for this detail.

The current state is that we use this mode only for debugging

Understood.

For editors like Vim, Emacs, etc, where users end up starting multiple instances on the same machine having a single instance of gopls will be a big win when it avoiding waiting for (pre-)warming of the imports cache (#34115).

Given that, do you have plans to fully support this mode?

[ianthehat]

Yes, but I have no time to do anything about it right now.
It is also valuable for people that want to run gopls from the command line if it could reuse the cached results already sitting behind their editor, or from a previous command.
Doing this well you really want to have a discovery protocol though, having to specify -remote on every command would be painful.
Ideally I would like to delete the -remote flag in the future and only support a discovery protocol!

[myitcv]

Yes, but I have no time to do anything about it right now.

Ok, understood. My question was more geared towards ensuring this is something that is being considered as opposed to not.

[ianthehat]

I am just being really careful to make sure people do not think I will get round to it any time soon, I don't want someone waiting on it and getting frustrated that I am not doing it!

This is also an area where contributions would be welcome, although it would be a very high touch contribution as I have a lot to say about how it is done :)

[myitcv]

Just to add another motivating factor behind this change: having a single instance will amortise the additional cost of running staticcheck on large code bases each time gopls is opened.

[findleyr]

Just to update, since this was discussed in slack: I'm working on this now, and should have some progress soon.

[gopherbot]

Change https://golang.org/cl/214803 mentions this issue: internal/lsp: add server instance to debug info

[myitcv]

@findleyr is this a v1.0.0 target of v0.3.0? Not chasing... just noting that it being worked on now probably contradicts the current "unplanned" milestone 😄

[findleyr]

Oh, thanks. Yes, I think this can be added to the v1.0.0 milestone (the v0.3.0 feature set is pretty locked-down at the moment). I'll confirm with @stamblerre.

[gopherbot]

Change https://golang.org/cl/215739 mentions this issue: internal/lsp/cmd: add a test for client logging

[gopherbot]

Change https://golang.org/cl/215740 mentions this issue: internal/lsp: remove the Context argument from NewSession

[findleyr]

@myitcv you didn't miss anything: that configuration isn't yet exposed. In fact, that's the last change I have planned for this issue, at which point I think we should close this and open new, more specific issues as needed.

I'd put off exposing configuration for the automatic daemon because I wasn't sure of what I wanted to do if the daemon is already running, but with different configuration parameters.

For now I will expose just the following two pieces of configuration:
-remote.listen.timeout -- the listen.timeout that should be used by an auto-started remote
-remote.debug -- the debug address configured on the remote server

Since neither of these affect serving, I think I'm NOT going to hash them into the daemon path as I do for build ID. This means that you may ask for -remote.listen.timeout=0, but if the daemon is already currently running with the default value of 1m for -remote.listen.timeout, it will continue running with that value. You'll need to restart everything to pick up the new configuration.

It will be an error to set either of these flags without also setting -remote=auto.*.

I think that keeps things simple for now -- we can always revisit in the future if this causes problems (for example, a race between govim and ALE might cause unpredictable startup configuration).

[gopherbot]

Change https://golang.org/cl/222669 mentions this issue: internal/lsp/cmd: add a dumpsessions command

[gopherbot]

Change https://golang.org/cl/222670 mentions this issue: internal/lsp/lsprpc: expose configuration for auto-started daemon

[gopherbot]

Change https://golang.org/cl/222671 mentions this issue: internal/lsp: clean up on Shutdown even if not initialized

[gopherbot]

Change https://golang.org/cl/222717 mentions this issue: gopls/doc: update daemon.md and remove 'experimental' caveats

[zikaeroh]

One thing I've noticed after doing a couple of gopls upgrades with this enabled is that the unused process seems to use CPU when I change my project's files even when no session is attached (e.g. I ask VSC to restart the language server, I get a new gopls serve process as the build ID changed, then edit a file while watching htop).

Is gopls doing some sort of extra file watching? I had thought it was only using LSP's file watcher, but the fact that it's responding to edits for files it doesn't have "open" is strange.

[zikaeroh]

Another thing I thought of: how does gopls in this mode determine which go binary to use? I'm guessing it inherits PATH from whichever process created it, which might be wrong for future gopls invocations. For example, if I have more than one Go version, I might set PATH accordingly, or if I'm working on the go tree itself, I might set PATH to point to that to see how things behave.

[findleyr]

@zikaeroh that's interesting re:CPU during gopls upgrades. I would not expect that, I'll see if I can reproduce.

The gopls forwarder starts the daemon with the same path as the forwarder itself. In other words, the sidecar gopls process uses its own path when starting the daemon. We can't use gopls from PATH because may LSP clients bundle a pinned gopls binary in their directory.

[zikaeroh]

I didn't mean which gopls binary, I meant which actual go binary it'd use via go/packages, go mod tidy, etc. I start my editor with code . which means I get whichever go is in PATH when I launch which I assume is what gopls is using. If the long running gopls server is doing the same then if I change which go I have in my PATH, the server won't see that change.

[findleyr]

I didn't mean which gopls binary, I meant which actual go binary it'd use via go/packages

Ah, sorry I misread your comment. You are right, it's possible that a change to the go binary path won't affect the daemon. It will have indeed picked up the environment from which it was started.

Perhaps we should hash the go binary build ID into the daemon socket path, similarly to how we handle the gopls binary.

[zikaeroh]

Of course, I see this becoming a slippery slope as I think there are other variables like GOPACKAGESDRIVER which can affect things too; there's probably a slew of things the existing hashing memoization takes as constant for convenience. Not to say that there isn't a line in the sand you can stop hashing at, of course... 🙂

[zikaeroh]

Now that this issue is closed, should I be filing new issues for the go binary and CPU usage items? Or are you already tracking them somewhere?

[findleyr]

I'm in the process of filing them now! You're too fast!