merge to stable-23-3 (#265)

* Count broken partitions instead of broken disks for placement groups [NBS] Сенсор PlacementGroupsWithRecentlyBrokenTwoOrMoreDisks неправильно работает с partition placement-группами. Он загорается, если ломается 2 и более дисков, но это могут быть диски в 1 партишне, что не страшно. Данный PR меняет логику для partition-placement групп - алерт будет загораться, если сломалось 2 и более партишнов. Для spread-placement групп каждый диск считается за партишн и для них логика не изменилась. * NBSNEBIUS-86: impld garbage FreshDeviceIds filtration in TVolumeState (#206) * NBSNEBIUS-86: not saying that replication is in progress on volume monpage if volume config contains only garbage FreshDeviceIds * NBSNEBIUS-86: moved FreshDeviceIds filtration to TVolumeState::Reset and removed it from the monpage code and removed the corresponding check from TMirrorPartitionState * NBSNEBIUS-86: rendering raw volume config (pretty-printed proto) on volume monpage (#220) * NBSNEBIUS-86: rendering raw volume config (pretty-printed proto) on volume monpage * NBSNEBIUS-86: rendering raw volume config (pretty-printed proto) on volume monpage * NBSNEBIUS-86: TDiskConfigs in DiskRegistry now store event history, history is displayed on DiskRegistry monpage, currently only ReplaceDevice events are added to history - will add other stuff in the next PRs (#233) * NBSNEBIUS-86: writing migration and replication-related events to disk history (#252) * NBSNEBIUS-86: fixed stupid bug: FilteredFreshDeviceIds filtration should not drop all FreshDeviceIds for volumes created after 2023-08-30 (#250) * NBSNEBIUS-86: fixed stupid bug: FilteredFreshDeviceIds filtration should not drop all FreshDeviceIds for volumes created after 2023-08-30 * NBSNEBIUS-86: moved FreshDeviceIds filtration to a separate func --------- Co-authored-by: dvrazumov <[email protected]>
ydb-platform · Jan 26, 2024 · 6e384b5 · 6e384b5
1 parent d71b1a7
commit 6e384b5
Show file tree

Hide file tree

Showing 19 changed files with 720 additions and 142 deletions.
diff --git a/cloud/blockstore/libs/storage/disk_registry/disk_registry_actor_mark_replacement_device.cpp b/cloud/blockstore/libs/storage/disk_registry/disk_registry_actor_mark_replacement_device.cpp
@@ -56,6 +56,7 @@ void TDiskRegistryActor::ExecuteMarkReplacementDevice(
     TDiskRegistryDatabase db(tx.DB);
 
     args.Error = State->MarkReplacementDevice(
+        ctx.Now(),
         db,
         args.DiskId,
         args.DeviceId,

diff --git a/cloud/blockstore/libs/storage/disk_registry/disk_registry_actor_monitoring.cpp b/cloud/blockstore/libs/storage/disk_registry/disk_registry_actor_monitoring.cpp
@@ -710,6 +710,33 @@ void TDiskRegistryActor::RenderDiskHtmlInfo(
         }
 
         GenerateVolumeActionsJS(out);
+
+        TAG(TH3) {
+            out << "History";
+        }
+
+        TABLE_SORTABLE_CLASS("table table-bordered") {
+            TABLEHEAD() {
+                TABLER() {
+                    TABLEH() { out << "Timestamp"; }
+                    TABLEH() { out << "Message"; }
+                }
+
+                for (const auto& hi: info.History) {
+                    TABLER() {
+                        TABLED() {
+                            out << TInstant::MicroSeconds(hi.GetTimestamp())
+                                << " (" << hi.GetTimestamp() << ")";
+                        }
+                        TABLED() {
+                            PRE() {
+                                out << hi.GetMessage();
+                            }
+                        }
+                    }
+                }
+            }
+        }
     }
 }
 
@@ -1424,11 +1451,11 @@ void TDiskRegistryActor::RenderPlacementGroupList(
                         auto it = brokenGroups.find(x.first);
 
                         if (it != brokenGroups.end()) {
-                            if (it->second.RecentlyBrokenDiskCount == 1) {
+                            if (it->second.Recently.GetBrokenPartitionsCount() == 1) {
                                 out << "<font color=yellow>&#9632;&nbsp;</font>";
                             }
 
-                            if (it->second.RecentlyBrokenDiskCount > 1) {
+                            if (it->second.Recently.GetBrokenPartitionsCount() > 1) {
                                 out << "<font color=red>&#9632;&nbsp;</font>";
                             }
                         }

diff --git a/cloud/blockstore/libs/storage/disk_registry/disk_registry_self_counters.cpp b/cloud/blockstore/libs/storage/disk_registry/disk_registry_self_counters.cpp
@@ -50,6 +50,14 @@ void TDiskRegistrySelfCounters::Init(
         counters->GetCounter("PlacementGroupsWithBrokenSingleDisk");
     PlacementGroupsWithBrokenTwoOrMoreDisks =
         counters->GetCounter("PlacementGroupsWithBrokenTwoOrMoreDisks");
+    PlacementGroupsWithRecentlyBrokenSinglePartition =
+        counters->GetCounter("PlacementGroupsWithRecentlyBrokenSinglePartition");
+    PlacementGroupsWithRecentlyBrokenTwoOrMorePartitions =
+        counters->GetCounter("PlacementGroupsWithRecentlyBrokenTwoOrMorePartitions");
+    PlacementGroupsWithBrokenSinglePartition =
+        counters->GetCounter("PlacementGroupsWithBrokenSinglePartition");
+    PlacementGroupsWithBrokenTwoOrMorePartitions =
+        counters->GetCounter("PlacementGroupsWithBrokenTwoOrMorePartitions");
     MeanTimeBetweenFailures =
         counters->GetCounter("MeanTimeBetweenFailures");
     AutomaticallyReplacedDevices =

diff --git a/cloud/blockstore/libs/storage/disk_registry/disk_registry_self_counters.h b/cloud/blockstore/libs/storage/disk_registry/disk_registry_self_counters.h
@@ -68,10 +68,17 @@ struct TDiskRegistrySelfCounters
     TCounterPtr Mirror3DisksMinus1;
     TCounterPtr Mirror3DisksMinus2;
     TCounterPtr Mirror3DisksMinus3;
+    // TODO(dvrazumov): "*Disk*" counters are replaced with "*Partitions" counters.
+    // They are left for compatibility and should be removed later (NBSNEBIUS-26)
     TCounterPtr PlacementGroupsWithRecentlyBrokenSingleDisk;
     TCounterPtr PlacementGroupsWithRecentlyBrokenTwoOrMoreDisks;
     TCounterPtr PlacementGroupsWithBrokenSingleDisk;
     TCounterPtr PlacementGroupsWithBrokenTwoOrMoreDisks;
+    // remove above ^^^
+    TCounterPtr PlacementGroupsWithRecentlyBrokenSinglePartition;
+    TCounterPtr PlacementGroupsWithRecentlyBrokenTwoOrMorePartitions;
+    TCounterPtr PlacementGroupsWithBrokenSinglePartition;
+    TCounterPtr PlacementGroupsWithBrokenTwoOrMorePartitions;
     TCounterPtr MeanTimeBetweenFailures;
     TCounterPtr AutomaticallyReplacedDevices;