Potential race condition between provision and delete

[gberche-orange]

It might be possible to have race condition between a provision request and a deprovision requests, resulting in dangling service instances in concourse that appear deleted in CF

Ways to address this:

• add mutex between provision and deprovision
• reduce race condition probability
  • by setting a timeout in git lower than the CC timeout.
    • plus keeping some headroom for potential git retries in between

See JGit Transport.setTimeout() number of seconds to wait (with no data transfer occurring) before aborting an IO read or write operation with this remote. and TransportHttp.java

[gberche-orange]

Setting JGit (connect + read timeout) timeout to 55s, and retry max delay to 50s. Hoping that in case of gitlab hangs, a single attempt will be made.