[Type 3] Audit thread-safety or other concurrency

[septatrix]

Summary

Currently it is not very clear if this library is thread-safe or not. According to https://github.com/psf/requests/issues/2766 a single session per thread is preferred so should we create multiple Server instances for each thread?

Alternatively I would be very interested in an asyncio Version of this library using httpx or a similar library. This is generally the direction in which Python is now also moving and a lot easier to reason about than multithreaded concurrency. Furthermore it would lead to better code style as all the hidden IO which is currently performed would then be eliminated.

[dxdc]

Ran into a similar issue, and as a result, I am quite sure the library is not thread safe.

These kinds of calls were causing race conditions for me (and as a result, giving me PDFs with the wrong filenames). I decided to use a semaphore for these calls in the meantime.

To reproduce, I was using this code with ThreadPoolExecutor, and specifying different filters (via pdf_request_options) for the same view. E.g.,

server.views.populate_pdf(view, pdf_request_options)
return view.pdf

You can also see the problem in a different way with the latest version of the library and the logging module, e.g., I can get multiple such calls in a row because this blocking state is not shared across threads.

2023-04-25 21:12:20,725 [DEBUG] None
None
2023-04-25 21:12:20,726 [DEBUG] None
None
2023-04-25 21:12:20,727 [DEBUG] None
[21:12:20] Async request failed: retrying
2023-04-25 21:12:20,727 [DEBUG] [21:12:20] Async request failed: retrying
[21:12:20] Async request failed: retrying
2023-04-25 21:12:20,727 [DEBUG] [21:12:20] Async request failed: retrying
[21:12:20] Async request failed: retrying
2023-04-25 21:12:20,727 [DEBUG] [21:12:20] Async request failed: retrying
[21:12:20] Begin blocking request to https://tableauserver.com/api/3.14/sites/xxxxx-site-1/views/xxxxx-view-2/pdf
2023-04-25 21:12:20,727 [DEBUG] [21:12:20] Begin blocking request to https://tableauserver.com/api/3.14/sites/xxxxx-site-1/views/xxxxx-view-2/pdf
[21:12:20] Begin blocking request to https://tableauserver.com/api/3.14/sites/xxxxx-site-1/views/xxxxx-view-2/pdf
2023-04-25 21:12:20,728 [DEBUG] [21:12:20] Begin blocking request to https://tableauserver.com/api/3.14/sites/xxxxx-site-1/views/xxxxx-view-2/pdf
[21:12:20] Begin blocking request to https://tableauserver.com/api/3.14/sites/xxxxx-site-1/views/xxxxx-view-2/pdf
2023-04-25 21:12:20,728 [DEBUG] [21:12:20] Begin blocking request to https://tableauserver.com/api/3.14/sites/xxxxx-site-1/views/xxxxx-view-2/pdf
[21:12:21] Call finished
2023-04-25 21:12:21,221 [DEBUG] [21:12:21] Call finished
[21:12:21] Request complete
2023-04-25 21:12:21,221 [DEBUG] [21:12:21] Request complete
[21:12:21] Call finished
2023-04-25 21:12:21,230 [DEBUG] [21:12:21] Call finished
[21:12:21] Request complete
2023-04-25 21:12:21,230 [DEBUG] [21:12:21] Request complete
Response status: <Response [200]>
2023-04-25 21:12:21,230 [DEBUG] Response status: <Response [200]>

This problem appears to extend to the Endpoint class itself, as rewriting _blocking_request and send_request_while_show_progress_threaded alone did not fix the problem.

https://github.com/tableau/server-client-python/blob/307d8a20a30f32c1ce615cca7c6a78b9b9bff081/tableauserverclient/server/endpoint/endpoint.py#L79-L95