tree 1cd5140efce60775f7f13e9c591296e24079089a
parent 44fe7765224fefbac8bc0a78ac0da1e4012a7266
author Sanil Jain <sanil.jain15@gmail.com> 1598295566 -0700
committer GitHub <noreply@github.com> 1598295566 -0700
gpgsig -----BEGIN PGP SIGNATURE-----
 
 wsBcBAABCAAQBQJfRA4OCRBK7hj4Ov3rIwAAdHIIAEm1CGzUpbR1czSEsOn88JO5
 XrAgewJptAw5Y2slwyVgEDlZl13s/wiC0r2WWnEiyJyNFwMI+S3hUE/YgK2DyXzq
 vaZYISEM9gLEAWNgX7h+/bkgPguE3LcRiO9XRpf8wfXycZRz3iQNKu6moKC0lRlz
 Wg1Gq5FRlXpmoHAZD26pAaNrkQ44333cRnkjN107b3jMRUqcyCV23GbHSJ2q8wyW
 MyOOkQFZTqbOjOB2r5f+Eh3MUXbRqC6BpEkIN72/fnMFo47O5vqChTCnhB9agUIa
 U7y7nhNtb7tyVFEhWBQsp+4v7dZ+nrhw049Xi/BGwl5coZp2fGiBdTT8W6QKWFw=
 =WIVK
 -----END PGP SIGNATURE-----
 

SAMZA-2579: Force restart feature for Container Placements (#1414)

Changes: The current restart ability for container placements works in the following way:

Tries to fetch resources on a host
Stops the active container if resources are accrued
Tried to start the container on host accrued
In production, we have seen the following observation at Linkedin

Some jobs are configured to use resources for the peak which leads to no headroom left on a host for requesting additional resources
This leads to restart requests failing due to not able to get resources on that host
A fix to this is to implement a force-restart utility , in this version we will stop the container first and then accrue resources. The upside being we will at least free up the resources on the host before issuing resource request, the downside being it will be a best-effort scenario to bring that container back up on that host

API Changes: Added new param values to destinationHost param for container placement request message

LAST_SEEN: Tries to restart a container on last seen host with RESERVE -> STOP -> MOVE policy

FORCE_RESTART_LAST_SEEN: Tries to restart a container on last seen host with STOP -> RESERVE -> MOVE policy