weizhouapache commented on code in PR #14386:
URL: https://github.com/apache/cloudstack/pull/14386#discussion_r4359371617
##########
plugins/hypervisors/kvm/src/main/java/com/cloud/hypervisor/kvm/resource/wrapper/LibvirtRestoreBackupCommandWrapper.java:
##########
@@ -381,6 +457,212 @@ private boolean
replaceBlockDeviceWithBackup(KVMStoragePoolManager storagePoolMg
return true;
}
+ /**
+ * Rebuild an RBD image by replaying a chain of {@code rbd export-diff}
streams into it: the
+ * chain's full anchor first, then each incremental in the order it was
taken.
+ *
+ * <p>RBD images are raw and cannot hold a backing-file reference, so —
unlike the qcow2 path —
+ * a diff carries no pointer to the one before it and the chain cannot be
flattened by
+ * {@code qemu-img}. Ceph applies the deltas natively instead: each diff
ends at the snapshot
+ * the backup was taken from, {@code import-diff} recreates that snapshot
on the target, and the
+ * next diff in the chain is validated against it. A truncated or
out-of-order diff is therefore
+ * rejected by Ceph rather than silently producing a half-restored
image.</p>
+ *
+ * <p>The image is recreated first. A full diff only carries the extents
that were allocated
+ * when it was taken, so any block still allocated in the current image
would otherwise show
+ * through the restore as stale data. Recreating the image also drops
every RBD snapshot it
+ * carried: a restore rewinds the volume wholesale, and snapshots of
content that is being
+ * replaced cannot be kept. The chain anchors this path takes are ours to
drop, but a snapshot
+ * taken outside it — a CloudStack volume snapshot kept on primary
storage, say — is not, so an
+ * image carrying one is refused rather than rewound.</p>
+ */
+ private boolean restoreRbdImageFromDiffChain(KVMStoragePoolManager
storagePoolMgr, PrimaryDataStoreTO volumePool,
+ String volumePath,
List<String> diffFiles, Long size) {
+ if (!Storage.StoragePoolType.RBD.equals(volumePool.getPoolType())) {
+ logger.error("Backup [{}] holds RBD diffs but volume [{}] is on {}
storage; it cannot be restored there.",
Review Comment:
for full backups, the backup is stored in qcow2 format, it can be restored
on other storage pools.
for the incremental backup, it is taken by "rbd" command, it can only be
restored to RBD pools.
##########
scripts/vm/hypervisor/kvm/nasbackup.sh:
##########
@@ -137,6 +138,282 @@ get_linstor_uuid_from_device() {
return 1
}
+# Globals populated by parse_rbd_opts() for the disk it was last called with.
+RBD_IMAGE="" # "<pool>/<image>"
+RBD_ARGS=() # connection arguments to pass to every rbd invocation
+RBD_KEYFILE="" # temporary cephx keyfile, removed by rbd_cleanup_keyfile
+# Prefix NASBackupProvider names every chain anchor with. Snapshots carrying
it belong to this
+# tool and no other, which is what makes the stale-anchor sweep safe.
+RBD_SNAPSHOT_PREFIX="csbackup-"
+
+rbd_cleanup_keyfile() {
+ if [[ -n "$RBD_KEYFILE" ]]; then
+ rm -f "$RBD_KEYFILE"
+ RBD_KEYFILE=""
+ fi
+}
+
+# Split an RBD disk string produced by KVMPhysicalDisk.RBDStringBuilder into
the image name and
+# the connection arguments the rbd CLI needs:
+#
rbd:<pool>/<image>:mon_host=<h1>\;<h2>:auth_client_required=cephx:id=<user>:key=<secret>:...
+# ':' and ';' are backslash-escaped inside mon_host so that option splitting
stays unambiguous;
+# they are masked before the split and restored afterwards.
+parse_rbd_opts() {
+ local s="${1#rbd:}"
+ s="${s//\\:/$'\x01'}"
+ s="${s//\\;/$'\x02'}"
+ local -a parts=()
+ IFS=':' read -ra parts <<< "$s"
+
+ rbd_cleanup_keyfile
+ RBD_IMAGE="${parts[0]}"
+ RBD_ARGS=()
+
+ local p monHost="" authId="" authKey=""
+ for p in "${parts[@]:1}"; do
+ case "$p" in
+ mon_host=*) monHost="${p#mon_host=}" ;;
+ id=*) authId="${p#id=}" ;;
+ key=*) authKey="${p#key=}" ;;
+ esac
+ done
+ monHost="${monHost//$'\x01'/:}" # host\:port -> host:port
+ monHost="${monHost//$'\x02'/,}" # monitor list -> the comma-separated form
rbd -m accepts
+
+ if [[ -n "$monHost" ]]; then
+ RBD_ARGS+=(-m "$monHost")
+ fi
+ if [[ -n "$authId" ]]; then
+ RBD_ARGS+=(--id "$authId")
+ if [[ -n "$authKey" ]]; then
+ RBD_KEYFILE=$(mktemp -t csbackup-ceph.XXXXXX)
+ chmod 600 "$RBD_KEYFILE"
+ printf '%s' "$authKey" > "$RBD_KEYFILE"
+ RBD_ARGS+=(--keyfile "$RBD_KEYFILE")
+ fi
+ fi
+}
+
+# Best-effort removal of snapshot $1 from each of the remaining disk
arguments. Used both to roll
+# back a half-created anchor and to drop the previous anchor once its delta
has been exported.
+rbd_remove_snapshots() {
+ local snapName="$1"
+ shift
+ local d
+ for d in "$@"; do
+ [[ -z "$d" ]] && continue
+ parse_rbd_opts "$d"
+ if ! rbd "${RBD_ARGS[@]}" snap rm "$RBD_IMAGE@$snapName" >> "$logFile"
2>&1; then
+ log -e "cleanup: failed to remove RBD snapshot $RBD_IMAGE@$snapName
(non-fatal)"
+ fi
+ done
+ rbd_cleanup_keyfile
+}
+
+# Remove every snapshot carrying the backup prefix except $1, across the
remaining disk arguments.
+# Only snapshots this tool created are ever touched: user and CloudStack
volume snapshots do not
+# use the prefix.
+rbd_sweep_stale_anchors() {
+ local keepSnap="$1"
+ shift
+ local d stale
+ for d in "$@"; do
+ [[ -z "$d" ]] && continue
+ parse_rbd_opts "$d"
+ while read -r stale; do
+ [[ -z "$stale" || "$stale" == "$keepSnap" ]] && continue
+ if ! rbd "${RBD_ARGS[@]}" snap rm "$RBD_IMAGE@$stale" >> "$logFile"
2>&1; then
+ log -e "cleanup: failed to remove stale RBD snapshot $RBD_IMAGE@$stale
(non-fatal)"
+ fi
+ done < <(
+ rbd "${RBD_ARGS[@]}" snap ls "$RBD_IMAGE" --format json 2>>"$logFile" |
python3 -c '
+import sys, json
+try:
+ snaps = json.load(sys.stdin)
+except Exception:
+ sys.exit(0)
+for s in snaps or []:
+ name = (s or {}).get("name") or ""
+ if name.startswith(sys.argv[1]):
+ print(name)
+' "$RBD_SNAPSHOT_PREFIX" 2>/dev/null || true
+ )
+ done
+ rbd_cleanup_keyfile
+}
+
+rbd_snapshot_exists() {
+ local snapName="$1"
+ rbd "${RBD_ARGS[@]}" snap ls "$RBD_IMAGE" --format json 2>>"$logFile" |
python3 -c '
+import sys, json
+try:
+ snaps = json.load(sys.stdin)
+except Exception:
+ sys.exit(1)
+sys.exit(0 if any((s or {}).get("name") == sys.argv[1] for s in snaps or [])
else 1)
+' "$snapName"
+}
+
+# Incremental backup of Ceph/RBD volumes.
+#
+# RBD images are raw, so they have nowhere to persist a QEMU dirty bitmap and
cannot take part in
+# the libvirt-checkpoint scheme backup_running_vm uses. Ceph provides the
equivalent natively: an
+# `rbd snap` per cycle as the anchor, and `rbd export-diff --from-snap` to
stream out exactly the
+# extents that changed between two anchors (computed inside Ceph from the
image object-map).
+#
+# Because the snapshot lives in Ceph rather than in the QEMU process, this
works identically for
+# running and stopped VMs, and survives a live migration of the VM between
hosts.
+#
+# Only the newest anchor is kept on each image: once this run's delta is
exported, the parent
+# snapshot is removed. A full additionally sweeps every other snapshot
carrying the backup prefix,
+# so anchors abandoned by a deleted chain, an unassigned offering or a restore
cannot accumulate.
+backup_rbd_vm() {
+ if [[ -z "$DISK_PATHS" ]]; then
+ echo "RBD backup mode requires the volume paths (-d|--diskpaths)"
+ exit 1
+ fi
+ if [[ -z "$BITMAP_NEW" ]]; then
+ echo "RBD backup mode requires --bitmap-new (the snapshot to create)"
+ exit 1
+ fi
+
+ local effective_mode="$MODE"
+ case "$effective_mode" in
+ full|incremental)
+ ;;
+ *)
+ echo "Mode $effective_mode is not supported by the rbd method"
+ cleanup
+ exit 1
+ ;;
+ esac
+ local -a disks=()
+ IFS=',' read -ra disks <<< "$DISK_PATHS"
+
+ local d
+ for d in "${disks[@]}"; do
+ if [[ "$d" != rbd:* ]]; then
+ echo "RBD backup mode was requested but disk $d is not an RBD volume"
+ exit 1
+ fi
+ done
+
+ # The parent anchor must still be present on EVERY disk. A volume migration
or an out-of-band
+ # snapshot purge can drop it; fall back to a full rather than letting
export-diff fail. Checked
+ # before anything is mounted or created so the fallback costs nothing.
+ if [[ "$effective_mode" == "incremental" ]]; then
+ for d in "${disks[@]}"; do
+ parse_rbd_opts "$d"
+ if ! rbd_snapshot_exists "$BITMAP_PARENT"; then
+ log -e "rbd incremental: parent snapshot $BITMAP_PARENT missing on
$RBD_IMAGE — falling back to full"
Review Comment:
done
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
