306-300 Sample Questions & Answers

pcs constraint location WebSite avoids node3 creates a location constraint with the default score INFINITY, i.e. a -INFINITY preference for node3; Red Hat: "A value of INFINITY for score in a pcs constraint location rsc avoids command indicates that the resource will never run on that node, even if no other node is available." pcs resource ban has no --temporary option (time-limited bans use lifetime=), crm_resource --move --node node3 creates a constraint that moves WebSite TO node3, and pcs order constraints take resources (not nodes) and accept kind/symmetrical options rather than a score.

First, the failed logical partition must be removed from the array configuration. Then, the new partition (assuming it has been partitioned correctly) is added back. The array will then begin the rebuild process.

Marking the OSD as 'out' instructs the CRUSH algorithm to re-map the Placement Groups (PGs) residing on that OSD to other OSDs. The cluster will begin backfilling data. Once the cluster is healthy (active+clean), the OSD can be safely stopped and removed.

LVS-TUN encapsulates the IP packet within another IP packet (IPIP). This allows the real servers to be on different networks or even across the internet, as long as they can decapsulate the packet and reply directly to the client (or via a gateway). LVS-NAT requires the load balancer to be the gateway. LVS-DR requires the load balancer and real servers to be on the same L2 segment.

A Replicated volume copies files to multiple bricks. With 4 servers and a replica count of 4 (or 2x2 distributed-replicated), read performance improves because clients can read from any copy. It provides redundancy.

Manual split-brain recovery (DRBD 9 User's Guide): on the split-brain victim (srv2, whose changes are discarded) run drbdadm disconnect r0, drbdadm secondary r0 and drbdadm connect --discard-my-data r0; on the survivor (srv1), if its connection state is also StandAlone, run drbdadm disconnect r0 and drbdadm connect r0 (if it is already Connecting it reconnects automatically). On connection srv2 becomes SyncTarget and only its local modifications are rolled back and replaced by srv1's data - no full resync. drbdadm invalidate on srv1 would replace the GOOD data with srv2's, taking srv2 down/up and forcing srv1 to Primary leaves the divergent data in place (split brain is detected again at the next handshake), and verify only marks differing blocks as out-of-sync.

Both commands create a 100 GiB thin pool named mythinpool in vg0. lvcreate --type thin-pool -L 100G -n mythinpool vg0 is the single-step form documented in lvmthin(7) (lvcreate --type thin-pool -L LargeSize -n ThinPoolLV VG). With -T|--thin, a size and no virtual size, lvcreate also creates a thin pool (--type thin-pool is implied), and the pool name can be given together with the VG name — the RHEL 7 LVM guide uses exactly lvcreate -L 100M -T vg001/mythinpool. vgcreate has no thin-pool option, and lvconvert --type thin-pool converts an existing LV into a pool and has no -L size option.

Both commands create a 100 GiB thin pool named mythinpool in vg0. lvcreate --type thin-pool -L 100G -n mythinpool vg0 is the single-step form documented in lvmthin(7) (lvcreate --type thin-pool -L LargeSize -n ThinPoolLV VG). With -T|--thin, a size and no virtual size, lvcreate also creates a thin pool (--type thin-pool is implied), and the pool name can be given together with the VG name — the RHEL 7 LVM guide uses exactly lvcreate -L 100M -T vg001/mythinpool. vgcreate has no thin-pool option, and lvconvert --type thin-pool converts an existing LV into a pool and has no -L size option.

A value of 100ms is the standard recommendation for miimon. It provides a good balance between fast detection (0.1 seconds) and low system overhead. The LACP standard also defines 'fast' and 'slow' rates for LACP PDUs, but miimon checks the carrier state locally.

STONITH (Shoot The Other Node In The Head) is a fencing implementation. It physically isolates a node (usually by cutting power via IPMI/PDU) to prevent it from writing to shared storage during a split-brain or malfunction.

Protocol A is asynchronous replication: a write is complete on the Primary as soon as it has reached the local disk and the replication packet has been placed in the local TCP send buffer. It gives the lowest write latency on long, high-latency links (it is the protocol typically used for long-distance replication, often together with DRBD Proxy), but after a forced failover the most recent writes may be lost. Protocol B (memory-synchronous) waits until the packet has reached the peer and only loses writes if both nodes lose power and the Primary's storage is destroyed at the same time; Protocol C (synchronous, the default) waits for the remote disk write. There is no Protocol D.