[WIP] Bypass http call if the node is local

[dharakk]

#13937
If a global config is set and the node-uri matches the current
node, then use the task manager directly for fetching exchange data and
doing task calls. That is bypass the taskResource if these conditions
are satisfied.

In addition, the backoff changed such that the local mode
never undergoes a backoff. It's a bit hacky but does the job.

Can be enabled (default is disabled) by setting
task.bypass-http-for-local=true

In the low latency one-node setup (designed by @agrawaldevesh ), large sum of latency stemmed from http calls on the local host, this change helps shave of that latency piece

== RELEASE NOTES ==

General Changes
* Bypass http call if the node is local with global config

[highker]

@dharakk, I just briefly skimmed through the PR and here are some high-level comments:

• having a communication tunnel bypassing network is ok to proceed
• we need better abstraction so that the local communication has nothing to do with HttpClient. Support internal communication with thrift #13894 shows an abstraction of how we can leverage it
• Given Support internal communication with thrift #13894 is there, could you rebase on top of it or wait till it's get reviewed (cc: @wenleix)? Because that patch has already provided you the abstraction. Be aware that the high-level principal of how exchange sink/source should be designed should be orthogonal from the detailed implementation (http, http2, thrift, grpc, or local function calls). A good abstraction is needed to achieve that.

[dharakk]

@dharakk, I just briefly skimmed through the PR and here are some high-level comments:

• having a communication tunnel bypassing network is ok to proceed
• we need better abstraction so that the local communication has nothing to do with HttpClient. Support internal communication with thrift #13894 shows an abstraction of how we can leverage it
• Given Support internal communication with thrift #13894 is there, could you rebase on top of it or wait till it's get reviewed (cc: @wenleix)? Because that patch has already provided you the abstraction. Be aware that the high-level principal of how exchange sink/source should be designed should be orthogonal from the detailed implementation (http, http2, thrift, grpc, or local function calls). A good abstraction is needed to achieve that.

Thanks @highker for the review. Having an abstraction is definitely the cleaner approach. I will rebase on top of #13894 and update

[highker]

As discussed, let's rebase this on top of #13894 with proper abstraction.

[stale]

This pull request has been automatically marked as stale because it has not had recent activity. If you'd still like this PR merged, please comment on the task, make sure you've addressed reviewer comments, and rebase on the latest master. Thank you for your contributions!