Validate torchserve process in pid file before starting

[rohithkrn]

Description

Fixes #1862 - There's an edge case where if .model_server.pid file is not deleted when torchserve stops abruptly and another process with same pid is started before torchserve is attempted to start again at which point it falsely assumes that the new pid of another process is torchserve process and exits with message "TorchServe is already running, please use torchserve --stop to stop TorchServe." This additional check validates that the process is actually a torchserve process and if not it will start torchserve.
Testing:

• Usual path - verified that torchserve start and stop work as before
• Edge case path - manually changed the pid file to replace pid with another process and verified that torchserve started without complaining that torchserve is already running.

[codecov]

Codecov Report

Merging #1866 (65ed2a0) into master (9217c6a) will decrease coverage by 0.13%.
The diff coverage is 0.00%.

❗ Current head 65ed2a0 differs from pull request most recent head e09ad75. Consider uploading reports for the commit e09ad75 to get more accurate results

@@            Coverage Diff             @@
##           master    #1866      +/-   ##
==========================================
- Coverage   45.13%   44.99%   -0.14%     
==========================================
  Files          63       63              
  Lines        2601     2609       +8     
  Branches       60       60              
==========================================
  Hits         1174     1174              
- Misses       1427     1435       +8     

Impacted Files	Coverage Δ
ts/model_server.py	0.00% <0.00%> (ø)

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[agunapal]

looks good

[msaroufim]

Could you please add some detail as to how you verified this fix worked?

[rohithkrn]

Could you please add some detail as to how you verified this fix worked?

@msaroufim updated the PR description to reflect the details

[khelkun]

Just to let you know that I have the same kind of issue on Windows server 2019, with TorchServe 0.7.1.

From Anaconda Prompt (ran as admninistrator), I run torchserve --start ..., everything goes fine including the inference test on the served model. I stop the torchserve --start ... command with CTRL+C.

I guess the SIGINT is not catched by torchserve.exe on Windows to delete the .model_server.pid from %APP_DATA%\Local\Temp\1\ so I have to delete it manually before running the next torchserve --start ... command.