[PATCH v3 05/24] qemu: Cancel storage migration in parallel

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Instead of cancelling disk mirrors sequentially, let's just call
block-job-cancel for all migrating disks and then wait until all
disappear.

In case we cancel disk mirrors at the end of successful migration we
also need to check all block jobs completed successfully. Otherwise we
have to abort the migration.

Signed-off-by: Jiri Denemark <jdenemar@xxxxxxxxxx>
---

Notes:
    Version 3:
    - new patch (separated from "qemu: Use domain condition for
      synchronous block jobs")
    - get rid of bool *failed parameter in qemuMigrationDriveMirrorCancelled

 src/qemu/qemu_migration.c | 196 +++++++++++++++++++++++++++++++---------------
 1 file changed, 134 insertions(+), 62 deletions(-)

diff --git a/src/qemu/qemu_migration.c b/src/qemu/qemu_migration.c
index 83d6c22..11504eb 100644
--- a/src/qemu/qemu_migration.c
+++ b/src/qemu/qemu_migration.c
@@ -1769,76 +1769,122 @@ qemuMigrationDriveMirrorReady(virQEMUDriverPtr driver,
 }
 
 
-/**
- * qemuMigrationCancelOneDriveMirror:
- * @driver: qemu driver
- * @vm: domain
+/*
+ * If @check is true, the function will report an error and return a different
+ * code in case a block job fails. This way we can properly abort migration in
+ * case some block jobs failed once all memory has already been transferred.
  *
- * Cancel all drive-mirrors started by qemuMigrationDriveMirror.
- * Any pending block job events for the mirrored disks will be
- * processed.
- *
- * Returns 0 on success, -1 otherwise.
+ * Returns 1 if all mirrors are gone,
+ *         0 if some mirrors are still active,
+ *         -1 some mirrors failed but some are still active,
+ *         -2 all mirrors are gone but some of them failed.
  */
 static int
-qemuMigrationCancelOneDriveMirror(virQEMUDriverPtr driver,
+qemuMigrationDriveMirrorCancelled(virQEMUDriverPtr driver,
                                   virDomainObjPtr vm,
-                                  virDomainDiskDefPtr disk)
+                                  bool check)
 {
-    qemuDomainObjPrivatePtr priv = vm->privateData;
-    char *diskAlias = NULL;
-    int ret = -1;
+    size_t i;
+    size_t active = 0;
+    int status;
+    bool failed = false;
 
-    /* No need to cancel if mirror already aborted */
-    if (disk->mirrorState == VIR_DOMAIN_DISK_MIRROR_STATE_ABORT) {
-        ret = 0;
-    } else {
-        virConnectDomainEventBlockJobStatus status = -1;
+    for (i = 0; i < vm->def->ndisks; i++) {
+        virDomainDiskDefPtr disk = vm->def->disks[i];
+        qemuDomainDiskPrivatePtr diskPriv = QEMU_DOMAIN_DISK_PRIVATE(disk);
 
-        if (virAsprintf(&diskAlias, "%s%s",
-                        QEMU_DRIVE_HOST_PREFIX, disk->info.alias) < 0)
-            goto cleanup;
+        if (!diskPriv->migrating)
+            continue;
 
-        if (qemuDomainObjEnterMonitorAsync(driver, vm,
-                                           QEMU_ASYNC_JOB_MIGRATION_OUT) < 0)
-            goto endjob;
-        ret = qemuMonitorBlockJobCancel(priv->mon, diskAlias, true);
-        if (qemuDomainObjExitMonitor(driver, vm) < 0)
-            goto endjob;
-
-        if (ret < 0) {
-            virDomainBlockJobInfo info;
-
-            /* block-job-cancel can fail if QEMU simultaneously
-             * aborted the job; probe for it again to detect this */
-            if (qemuMonitorBlockJobInfo(priv->mon, diskAlias,
-                                        &info, NULL) == 0) {
-                ret = 0;
-            } else {
+        status = qemuBlockJobUpdate(driver, vm, disk);
+        switch (status) {
+        case VIR_DOMAIN_BLOCK_JOB_FAILED:
+            if (check) {
                 virReportError(VIR_ERR_OPERATION_FAILED,
-                               _("could not cancel migration of disk %s"),
+                               _("migration of disk %s failed"),
                                disk->dst);
+                failed = true;
             }
+            /* fallthrough */
+        case VIR_DOMAIN_BLOCK_JOB_CANCELED:
+        case VIR_DOMAIN_BLOCK_JOB_COMPLETED:
+            qemuBlockJobSyncEnd(driver, vm, disk);
+            diskPriv->migrating = false;
+            break;
 
-            goto endjob;
+        default:
+            active++;
         }
+    }
 
-        /* Mirror may become ready before cancellation takes
-         * effect; loop if we get that event first */
-        while (1) {
-            status = qemuBlockJobUpdate(driver, vm, disk);
-            if (status != -1 && status != VIR_DOMAIN_BLOCK_JOB_READY)
-                break;
-            if ((ret = virDomainObjWait(vm)) < 0)
-                goto endjob;
+    if (failed) {
+        if (active) {
+            VIR_DEBUG("Some disk mirrors failed; still waiting for %zu "
+                      "disk mirrors to finish", active);
+            return -1;
+        } else {
+            VIR_DEBUG("All disk mirrors are gone; some of them failed");
+            return -2;
+        }
+    } else {
+        if (active) {
+            VIR_DEBUG("Waiting for %zu disk mirrors to finish", active);
+            return 0;
+        } else {
+            VIR_DEBUG("All disk mirrors are gone");
+            return 1;
         }
     }
+}
+
+
+/*
+ * Returns 0 on success,
+ *         1 when job is already completed or it failed and failNoJob is false,
+ *         -1 on error or when job failed and failNoJob is true.
+ */
+static int
+qemuMigrationCancelOneDriveMirror(virQEMUDriverPtr driver,
+                                  virDomainObjPtr vm,
+                                  virDomainDiskDefPtr disk,
+                                  bool failNoJob)
+{
+    qemuDomainObjPrivatePtr priv = vm->privateData;
+    char *diskAlias = NULL;
+    int ret = -1;
+    int status;
+    int rv;
+
+    status = qemuBlockJobUpdate(driver, vm, disk);
+    switch (status) {
+    case VIR_DOMAIN_BLOCK_JOB_FAILED:
+    case VIR_DOMAIN_BLOCK_JOB_CANCELED:
+        if (failNoJob) {
+            virReportError(VIR_ERR_OPERATION_FAILED,
+                           _("migration of disk %s failed"),
+                           disk->dst);
+            return -1;
+        }
+        return 1;
+
+    case VIR_DOMAIN_BLOCK_JOB_COMPLETED:
+        return 1;
+    }
+
+    if (virAsprintf(&diskAlias, "%s%s",
+                    QEMU_DRIVE_HOST_PREFIX, disk->info.alias) < 0)
+        return -1;
+
+    if (qemuDomainObjEnterMonitorAsync(driver, vm,
+                                       QEMU_ASYNC_JOB_MIGRATION_OUT) < 0)
+        goto cleanup;
+
+    rv = qemuMonitorBlockJobCancel(priv->mon, diskAlias, true);
 
- endjob:
-    qemuBlockJobSyncEnd(driver, vm, disk);
+    if (qemuDomainObjExitMonitor(driver, vm) < 0 || rv < 0)
+        goto cleanup;
 
-    if (disk->mirrorState == VIR_DOMAIN_DISK_MIRROR_STATE_ABORT)
-        disk->mirrorState = VIR_DOMAIN_DISK_MIRROR_STATE_NONE;
+    ret = 0;
 
  cleanup:
     VIR_FREE(diskAlias);
@@ -1850,6 +1896,7 @@ qemuMigrationCancelOneDriveMirror(virQEMUDriverPtr driver,
  * qemuMigrationCancelDriveMirror:
  * @driver: qemu driver
  * @vm: domain
+ * @check: if true report an error when some of the mirrors fails
  *
  * Cancel all drive-mirrors started by qemuMigrationDriveMirror.
  * Any pending block job events for the affected disks will be
@@ -1859,28 +1906,53 @@ qemuMigrationCancelOneDriveMirror(virQEMUDriverPtr driver,
  */
 static int
 qemuMigrationCancelDriveMirror(virQEMUDriverPtr driver,
-                               virDomainObjPtr vm)
+                               virDomainObjPtr vm,
+                               bool check)
 {
     virErrorPtr err = NULL;
-    int ret = 0;
+    int ret = -1;
     size_t i;
+    int rv;
+    bool failed = false;
+
+    VIR_DEBUG("Cancelling drive mirrors for domain %s", vm->def->name);
 
     for (i = 0; i < vm->def->ndisks; i++) {
         virDomainDiskDefPtr disk = vm->def->disks[i];
         qemuDomainDiskPrivatePtr diskPriv = QEMU_DOMAIN_DISK_PRIVATE(disk);
 
-        if (!diskPriv->migrating || !diskPriv->blockJobSync)
+        if (!diskPriv->migrating)
             continue;
 
-        if (qemuMigrationCancelOneDriveMirror(driver, vm, disk) < 0) {
-            ret = -1;
-            if (!err)
-                err = virSaveLastError();
+        rv = qemuMigrationCancelOneDriveMirror(driver, vm, disk, check);
+        if (rv != 0) {
+            if (rv < 0) {
+                if (!err)
+                    err = virSaveLastError();
+                failed = true;
+            }
+            qemuBlockJobSyncEnd(driver, vm, disk);
+            diskPriv->migrating = false;
+        }
+    }
+
+    while ((rv = qemuMigrationDriveMirrorCancelled(driver, vm, check)) != 1) {
+        if (rv < 0) {
+            failed = true;
+            if (rv == -2)
+                break;
         }
 
-        diskPriv->migrating = false;
+        if (failed && !err)
+            err = virSaveLastError();
+
+        if (virDomainObjWait(vm) < 0)
+            goto cleanup;
     }
 
+    ret = failed ? -1 : 0;
+
+ cleanup:
     if (err) {
         virSetError(err);
         virFreeError(err);
@@ -3532,7 +3604,7 @@ qemuMigrationConfirmPhase(virQEMUDriverPtr driver,
         virErrorPtr orig_err = virSaveLastError();
 
         /* cancel any outstanding NBD jobs */
-        qemuMigrationCancelDriveMirror(driver, vm);
+        qemuMigrationCancelDriveMirror(driver, vm, false);
 
         virSetError(orig_err);
         virFreeError(orig_err);
@@ -4111,7 +4183,7 @@ qemuMigrationRun(virQEMUDriverPtr driver,
 
     /* cancel any outstanding NBD jobs */
     if (mig && mig->nbd) {
-        if (qemuMigrationCancelDriveMirror(driver, vm) < 0)
+        if (qemuMigrationCancelDriveMirror(driver, vm, ret == 0) < 0)
             ret = -1;
     }
 
-- 
2.4.3

--
libvir-list mailing list
libvir-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/libvir-list




[Index of Archives]     [Virt Tools]     [Libvirt Users]     [Lib OS Info]     [Fedora Users]     [Fedora Desktop]     [Fedora SELinux]     [Big List of Linux Books]     [Yosemite News]     [KDE Users]     [Fedora Tools]