lsm/seaweedfs
1
0
Fork 0
mirror of https://github.com/seaweedfs/seaweedfs.git synced 2025-11-10 07:27:27 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Improve EC shards rebalancing logic across racks (#6270)

Browse Source 
Improve EC shards rebalancing logic across racks.

  - Favor target shards with less preexisting shards, to ensure a fair distribution.
  - Randomize selection when multiple possible target shards are available.
  - Add logic to account for replication settings when selecting target shards (currently disabled).
This commit is contained in:
Lisandro Pin
2024-11-21 17:46:24 +01:00
committed by GitHub
parent e56327e3b0
commit ca499de1cb
4 changed files with 274 additions and 45 deletions
Download Patch File Download Diff File Expand all files Collapse all files

84
weed/shell/command_ec_common_test.go
View File

@@ -10,12 +10,16 @@ import (
"github.com/seaweedfs/seaweedfs/weed/storage/super_block"
)
var (
topology1 = parseOutput(topoData)
topology2 = parseOutput(topoData2)
topologyEc = parseOutput(topoDataEc)
)
func TestEcDistribution(t *testing.T) {
topologyInfo := parseOutput(topoData)
// find out all volume servers with one slot left.
ecNodes, totalFreeEcSlots := collectEcVolumeServersByDc(topologyInfo, "")
ecNodes, totalFreeEcSlots := collectEcVolumeServersByDc(topology1, "")
sortEcNodesByFreeslotsDescending(ecNodes)
@@ -34,23 +38,20 @@ func TestEcDistribution(t *testing.T) {
}
func TestVolumeIdToReplicaPlacement(t *testing.T) {
topo1 := parseOutput(topoData)
topo2 := parseOutput(topoData2)
testCases := []struct {
topology *master_pb.TopologyInfo
vid string
want string
wantErr string
}{
{topo1, "", "", "failed to resolve replica placement for volume ID 0"},
{topo1, "0", "", "failed to resolve replica placement for volume ID 0"},
{topo1, "1", "100", ""},
{topo1, "296", "100", ""},
{topo2, "", "", "failed to resolve replica placement for volume ID 0"},
{topo2, "19012", "", "failed to resolve replica placement for volume ID 19012"},
{topo2, "6271", "002", ""},
{topo2, "17932", "002", ""},
{topology1, "", "", "failed to resolve replica placement for volume ID 0"},
{topology1, "0", "", "failed to resolve replica placement for volume ID 0"},
{topology1, "1", "100", ""},
{topology1, "296", "100", ""},
{topology2, "", "", "failed to resolve replica placement for volume ID 0"},
{topology2, "19012", "", "failed to resolve replica placement for volume ID 19012"},
{topology2, "6271", "002", ""},
{topology2, "17932", "002", ""},
}
for _, tc := range testCases {
@@ -59,29 +60,74 @@ func TestVolumeIdToReplicaPlacement(t *testing.T) {
got, gotErr := volumeIdToReplicaPlacement(vid, ecNodes)
if tc.wantErr == "" && gotErr != nil {
t.Errorf("expected no error for volume '%s', got '%s'", tc.vid, gotErr.Error())
t.Errorf("expected no error for volume %q, got %q", tc.vid, gotErr.Error())
continue
}
if tc.wantErr != "" {
if gotErr == nil {
t.Errorf("got no error for volume '%s', expected '%s'", tc.vid, tc.wantErr)
t.Errorf("got no error for volume %q, expected %q", tc.vid, tc.wantErr)
continue
}
if gotErr.Error() != tc.wantErr {
t.Errorf("expected error '%s' for volume '%s', got '%s'", tc.wantErr, tc.vid, gotErr.Error())
t.Errorf("expected error %q for volume %q, got %q", tc.wantErr, tc.vid, gotErr.Error())
continue
}
}
if got == nil {
if tc.want != "" {
t.Errorf("expected replica placement '%s' for volume '%s', got nil", tc.want, tc.vid)
t.Errorf("expected replica placement %q for volume %q, got nil", tc.want, tc.vid)
}
continue
}
want, _ := super_block.NewReplicaPlacementFromString(tc.want)
if !got.Equals(want) {
t.Errorf("got replica placement '%s' for volune '%s', want '%s'", got.String(), tc.vid, want.String())
t.Errorf("got replica placement %q for volune %q, want %q", got.String(), tc.vid, want.String())
}
}
}
func TestPickRackToBalanceShardsInto(t *testing.T) {
testCases := []struct {
topology *master_pb.TopologyInfo
vid string
wantOneOf []string
}{
// Non-EC volumes. We don't care about these, but the function should return all racks as a safeguard.
{topologyEc, "", []string{"rack1", "rack2", "rack3", "rack4", "rack5", "rack6"}},
{topologyEc, "6225", []string{"rack1", "rack2", "rack3", "rack4", "rack5", "rack6"}},
{topologyEc, "6226", []string{"rack1", "rack2", "rack3", "rack4", "rack5", "rack6"}},
{topologyEc, "6241", []string{"rack1", "rack2", "rack3", "rack4", "rack5", "rack6"}},
{topologyEc, "6242", []string{"rack1", "rack2", "rack3", "rack4", "rack5", "rack6"}},
// EC volumes.
{topologyEc, "9577", []string{"rack1", "rack2", "rack3"}},
{topologyEc, "10457", []string{"rack1"}},
{topologyEc, "12737", []string{"rack2"}},
{topologyEc, "14322", []string{"rack3"}},
}
for _, tc := range testCases {
vid, _ := needle.NewVolumeId(tc.vid)
ecNodes, _ := collectEcVolumeServersByDc(tc.topology, "")
racks := collectRacks(ecNodes)
locations := ecNodes
rackToShardCount := countShardsByRack(vid, locations)
averageShardsPerEcRack := ceilDivide(erasure_coding.TotalShardsCount, len(racks))
got := pickRackToBalanceShardsInto(racks, rackToShardCount, nil, averageShardsPerEcRack)
if string(got) == "" && len(tc.wantOneOf) == 0 {
continue
}
found := false
for _, want := range tc.wantOneOf {
if got := string(got); got == want {
found = true
break
}
}
if !(found) {
t.Errorf("expected one of %v for volume %q, got %q", tc.wantOneOf, tc.vid, got)
}
}
}