5283: Improve reliability of crunch-job output collation.
authorBrett Smith <brett@curoverse.com>
Fri, 27 Feb 2015 19:20:12 +0000 (14:20 -0500)
committerBrett Smith <brett@curoverse.com>
Fri, 27 Feb 2015 19:20:12 +0000 (14:20 -0500)
commiteab3fd07f8f2eea43a373454a91ba0ca3fc60a6b
tree48e5bb3eeb1d66fc70a9ef9d0143dd625b1254aa
parentb10c68ef2d5ec1418653594576e8d66b1df3453e
5283: Improve reliability of crunch-job output collation.

* Check the results of all pipe opens, exit statuses, and writes.
  Log any problems.
* Have fetch_block return undef when it encounters trouble, rather
  than dying.  create_output_collection already checks for this, so it
  effectively bubbles up the error.
* Retry all of the associated API calls.
* Kill the manifest creation pipe if we give up on it, per the TODO.

This probably won't resolve #5283, but hopefully these changes will
give us additional information to help diagnose the problem.
sdk/cli/bin/crunch-job