#2861 - sqlserver input plugin not working with Azure SQL

[regevbr]

#2861 - sqlserver input plugin not working with Azure SQL

• Signed CLA.
• Associated README.md updated.
• Has appropriate unit tests.

[deluxor]

I will test your PR later, and i'll post some feedback

[regevbr]

Thanks @deluxor

[regevbr]

@deluxor Im done - added a param "disable_parallel" that will cause all queries within a server to run one by one - this is for some reason needed by Azure SQL - probably due to DOS detection.
In addition, added a needed first simple query to sustain the session to the DB (denisenkom/go-mssqldb#217).
In addition, to save resources, each server will have a single connection open to it, instead of 1.
And finally, to resolve the missing sys views, added to the queries options to ignore those views if they don't exist
I replaced the mssql driver because of what @zensqlmonitor said in #2852

[deluxor]

Great, I'll give it a try and back to you soon.

[regevbr]

The new plugin is running in my production env for a few hours now and it works prefect

[danielnelson]

Is this because we are recreating a new connection pool every gather? We have an in progress PR to add this to postgres #2701, would this method work here?

[regevbr]

That was not the issue here, but it will make sense to implement the behavior here as well...
There waa a misuse of the connection as it was created per query, which i fixed, but having a global connection will make it better, but I think it belongs to a different pull request

[danielnelson]

Okay, so previously we would create multiple new connections per Gather. With this change we can use a single new connection per Gather?

I don't really want an option to control if queries are run in parallel. I'm not experienced with the go sql library but it would make more sense to me to be able to configure the connection pool size that is used for the life of the plugin.

[regevbr]

Yes, single connection per connection string per gathering. Currently every run of the gather method, a new connection is being open, if we want one connection it ia a different request.
As to the parallel queries, it is required to run the queries im sequential order for the DOS mechanism not to get triggered

[danielnelson]

I think adding this option as is leaks too much detail about the implementation. What if we added an option called max_connections? For now we could use a semaphore to limit the connections to this amount but in the future if we add support for a stored connection pool later it would still apply.

[regevbr]

Sounds good to me, I will fix the pull request

[danielnelson]

I believe this was addressed in #3618