Description
Right now the scheduler hardcodes "host" and "rack" limit constraints into submitted jobs. These require slave attributes to be set on the slave command-line, which means deploying aurora on a running mesos cluster needs a hard slave restart (changing slave attributes invalidates recovery metadata, meaning underlying tasks must be killed). In addition, "host" and "rack" might not accurately capture failure domains in other mesos deployments.
Make this constraint-set configurable at deploy-time, defaulting to empty.
Attachments
Issue Links
- duplicates
-
AURORA-174 injected scheduler constraints should be configurable
- Resolved
- is duplicated by
-
AURORA-212 Modify default rack limit to be something other than 1 (perhaps N/10)
- Resolved
-
AURORA-662 Remove Rack Limit Default
- Resolved
- relates to
-
AURORA-174 injected scheduler constraints should be configurable
- Resolved