Fixes
store: the Jobs page and the orphan check no longer read the whole deployment history on each refresh and detection cycle: their cost follows the number of jobs, however long the history grows. The upgrade reads the history once to fill a table of the latest deployment of each job (#315).store: the lists, newest first, no longer show rows of the same second oldest first, and the latest deployment of a job is no longer the older of two created within one second. The upgrade rewrites the stored timestamps once (#316).meta: anops_*key that doesn't apply to its job logs a WARN that names it and says why: a managed job's key on a hook, andnops_pre_hookornops_post_hookon a job that isn't managed (validation, #317).engine: a job without a Nomad deployment that Nomad places only in part no longer completes its deployment: it waits for the rest, up to the apply timeout. An allocation Nomad gives up on fails the deployment at once with its own failure, even while another part of the job waits for room. When the apply timeout runs out while no node has room for part of a job, its message names the cause (healthy, the apply timeout, #318).
Commits: v0.15.4...v0.15.5