Add telemetry for language server installs

[qubitron]

We should collect telemetry on language server download times and failures, so that we can become aware of issues.

I propose we add events that allow us to track the following:

• How many users attempt to download the language server
• How many succeed
• How many fail, and reasons for failure
• Amount of time taken for each of: download, unpack, and install
• Time taken to analyze workspace first time after install

[MikhailArkhipov]

We already report

    PYTHON_LANGUAGE_SERVER_DOWNLOADED,
    PYTHON_LANGUAGE_SERVER_ENABLED,
    PYTHON_LANGUAGE_SERVER_ERROR

The latter has error text attached

[d3r3kk]

@qubitron , just FYI, there isn't really an install phase per se. There is download, then extract - that is pretty much it.

As for the time to analyze the workspace for the first time, what is the question you are trying to answer here?
(I think perhaps that won't tell you too much useful information as it truly does depend on the size of that workspace? The users processor speed/IO capacity/etc... also factor in)

[qubitron]

@d3r3kk I'm interested to know on what the first time user experience looks like, for the language server, are people waiting seconds, minutes, or hours before they can start using it. Being able to break that down into download, extract, first time analysis analyzing will help us break that down further to know where any issues are.

I'm also interested to know in general how much time people wait for workspace analysis to complete on an ongoing basis, and how many people are getting stuck on the "analyzing workspace" issue #2297 (analysis starts but never completes).

[d3r3kk]

Ok, that's cool and makes sense. Generally speaking, how long do our users wait for analysis to complete, and how many folks are affected by any never-complete scenarios.

Would it be better/more preferrable then, to reword the last item in your list as:

• Time taken to analyze workspace every time.

...and then add:

• If analysis takes > N seconds, report a never-complete scenario.

[d3r3kk]

Alternatively, we could say:

• Time taken to analyze workspace for the first N times.

...to keep it a bit more reasonable perhaps.

[DonJayamanne]

Time taken to analyze workspace every time.
...and then add:
If analysis takes > N seconds, report a never-complete scenario.

I think this is better. From this we can capture the success rate for completion of analysis.