(Click to open topic with navigation)
Checkpointing records the state of a job, allowing for it to restart later without interruption to the job's execution. Checkpointing can be performed manually, as the result of triggers or events, or in conjunction with various QoS policies.
Moab's ability to checkpoint is dependent upon both the cluster's resource manager and operating system. In most cases, two types of checkpoint are enabled, including (1) checkpoint and continue and (2) checkpoint and terminate. While either checkpointing method can be activated using the mjobctl command, only the checkpoint and terminate type is used by internal scheduling and event managements facilities.
Checkpointing behavior can be configured on a per-resource manager basis using various attributes of the RMCFG parameter.
Related topics