Unauthorized Access error when signing out (either explicitly or at the end of a `with` block)
bcantoni opened this issue · 8 comments
Starting with v0.29 I notice many of my test scripts failing with an exception. It seems like a general problem rather than specific to one endpoint. Here's an example:
import tableauserverclient as TSC
tableau_auth = TSC.PersonalAccessTokenAuth(
"xxxxx",
"xxxxxxxxxx",
"",
)
server = TSC.Server("https://devplat.tableautest.com", use_server_version=True)
with server.auth.sign_in(tableau_auth):
all_wb, pagination_item = server.workbooks.get()
print("\nThere are {} workbooks: ".format(pagination_item.total_available))
for wb in all_wb:
print(wb.id, wb.name, wb.tags)The script succeeds (printing workbooks), but ends with an exception like this:
Traceback (most recent call last):
File "/Users/bcantoni/github/server-client-python/getdatasources.py", line 14, in <module>
with server.auth.sign_in(tableau_auth):
File "/Users/bcantoni/github/server-client-python/tableauserverclient/server/endpoint/auth_endpoint.py", line 27, in __exit__
self._callback()
File "/Users/bcantoni/github/server-client-python/tableauserverclient/server/endpoint/endpoint.py", line 291, in wrapper
return func(self, *args, **kwargs)
File "/Users/bcantoni/github/server-client-python/tableauserverclient/server/endpoint/auth_endpoint.py", line 85, in sign_out
self.post_request(url, "")
File "/Users/bcantoni/github/server-client-python/tableauserverclient/server/endpoint/endpoint.py", line 248, in post_request
return self._make_request(
File "/Users/bcantoni/github/server-client-python/tableauserverclient/server/endpoint/endpoint.py", line 165, in _make_request
self._check_status(server_response, url)
File "/Users/bcantoni/github/server-client-python/tableauserverclient/server/endpoint/endpoint.py", line 186, in _check_status
raise NotSignedInError(server_response.content, url)
tableauserverclient.server.endpoint.exceptions.NotSignedInError: (b'<?xml version=\'1.0\' encoding=\'UTF-8\'?><tsResponse xmlns="http://tableau.com/api" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://tableau.com/api https://help.tableau.com/samples/en-us/rest_api/ts-api_3_22.xsd"><error code="401002"><summary>Unauthorized Access</summary><detail>Invalid authentication credentials were provided.</detail></error></tsResponse>', 'https://10ax.online.tableau.com/api/3.22/auth/signout')
If I switch from using the with statement to the older style, it works fine:
import tableauserverclient as TSC
tableau_auth = TSC.PersonalAccessTokenAuth(
"xxxx",
"xxxxxxx",
"",
)
server = TSC.Server("https://devplat.tableautest.com", use_server_version=True)
server.auth.sign_in(tableau_auth)
all_wb, pagination_item = server.workbooks.get()
print("\nThere are {} workbooks: ".format(pagination_item.total_available))
for wb in all_wb:
print(wb.id, wb.name, wb.tags)Testing notes:
- I tested with same results on both Tableau Cloud and Server.
- This seems new to 0.29; switching back to 0.28 solves the issue.
- Git bisect points to #1300 as the point this was introduced
- In server/endpoint/endpoint.py, changing
seconds = 0.05back toseconds = 0seems to fix it (but I'll admit I don't totally follow the changes in that PR)
I think this is due to the server.auth.sign_out() which is automatically called at the end of the with-statement.
When you include the logout in your older style this error should come up aswell.
@bcantoni - Confirmed. We are seeing the same error message initiating refreshes on Tableau Cloud on the latest copy of tableauserverclient from within our instance of Databricks (running python). Explicitly installing 0.28 seems to fix the problem. (Thank you @mfalkenham).
NotSignedInError: (b'Unauthorized AccessInvalid authentication credentials were provided.', 'https://us-east-1.online.tableau.com/api/3.21/auth/signout')
Is there a decision from Tableau on if this will be fixed? Or should we just update our code to remove the "with" as noted above?
@BerndRos ahh thanks for pointing that out - I confirmed in my workaround calling sign_out does cause the same exception.
Another fast and dirty solution
import tableauserverclient as TSC
tableau_auth = TSC.PersonalAccessTokenAuth(
"xxxxx",
"xxxxxxxxxx",
"",
)
server = TSC.Server("https://devplat.tableautest.com", use_server_version=True)
server.auth.sign_out = lambda: None # Set sign_out() as a empty func
with server.auth.sign_in(tableau_auth):
all_wb, pagination_item = server.workbooks.get()
print("\nThere are {} workbooks: ".format(pagination_item.total_available))
for wb in all_wb:
print(wb.id, wb.name, wb.tags)This error is caused by the threading code repeating the signout call. I'm pretty sure that all of the four issues logged on 0.29 are caused the same way. I'm working on a fix.
Don't want to add any pressure but do you have a rough idea when the fix for this might be released? We're trying to plan our server upgrade and the release will help us figure out a good plan for it 😄
@TWeatherston @BobDu @gordonstrodel @BerndRos FYI the fix for this has been released now as part of v0.30 https://github.com/tableau/server-client-python/releases/tag/v0.30
@bcantoni , just saw this now. I'd like to give a bit of post-mortem info about #1300 and #1299 and this issue.
#1299 was an issue with only some versions of tableau-server, and I'd thought that the server was timing out while waiting to transfer another chunk in large hyper-files. My solution was to reduce the waiting times in order to catch the responses faster. At the beginning of the async request I'd shortened the wait from 1 s to 0.05 s. And in the loops of waiting from 10 s to 0.1 s.
This solved the issue of failed uploads with larger hyperfiles on older versions of tableau server, where we discovered the issue was Tableau-Server 2022.1.13. We were not able to reproduce the issue on 2022.1.16, so it seemed to depend on the Server version.
In hindsight, it looks like the async completion flag that was being used hadn't been working reliably. To find the exact cause of the issue, I'd suggest looking at the client's compatibility with multiple versions of tableau-server, as I suspect the new problem was introduced with an update of tableau-server and not with the client. Perhaps the timings for transactions were made tighter with the new version, reintroducing this issue, which makes sense for me that by further reducing the sleep time at the start of the transaction had solved the issue for you.
The client did have an overall issue with the async completion, so the new fix of removing async completely seems like a more robust solution.