choose switchover candidate based on lowest lag in MB

This commit is contained in:
Felix Kunde 2021-12-02 15:35:34 +01:00
parent d784c961b4
commit 4f87238fe0
2 changed files with 22 additions and 25 deletions

View File

@ -4,8 +4,8 @@ import (
"context"
"fmt"
"math/rand"
"sort"
"strconv"
"strings"
"time"
appsv1 "k8s.io/api/apps/v1"
@ -483,9 +483,8 @@ func (c *Cluster) recreatePods(pods []v1.Pod, switchoverCandidates []spec.Namesp
func (c *Cluster) getSwitchoverCandidate(master *v1.Pod) (spec.NamespacedName, error) {
var members []patroni.ClusterMember
candidates := make([]spec.NamespacedName, 0)
syncCandidates := make([]spec.NamespacedName, 0)
skipReasons := make([]string, 0)
candidates := make([]patroni.ClusterMember, 0)
syncCandidates := make([]patroni.ClusterMember, 0)
err := retryutil.Retry(1*time.Second, 5*time.Second,
func() (bool, error) {
@ -503,32 +502,24 @@ func (c *Cluster) getSwitchoverCandidate(master *v1.Pod) (spec.NamespacedName, e
}
for _, member := range members {
if member.LagInMB > 0 {
skipReasons = append(skipReasons, fmt.Sprintf("%s lags behind by %d MB", member.Name, member.LagInMB))
continue
}
if PostgresRole(member.Role) != Leader && PostgresRole(member.Role) != StandbyLeader && member.State == "running" {
candidates = append(candidates, spec.NamespacedName{Namespace: master.Namespace, Name: member.Name})
if PostgresRole(member.Role) != SyncStandby {
syncCandidates = append(syncCandidates, spec.NamespacedName{Namespace: master.Namespace, Name: member.Name})
candidates = append(candidates, member)
if PostgresRole(member.Role) == SyncStandby {
syncCandidates = append(syncCandidates, member)
}
}
}
if len(syncCandidates) > 0 {
return candidates[rand.Intn(len(syncCandidates))], nil
sort.Slice(syncCandidates, func(i, j int) bool { return syncCandidates[i].LagInMB < syncCandidates[j].LagInMB })
return spec.NamespacedName{Namespace: master.Namespace, Name: syncCandidates[0].Name}, nil
}
if len(candidates) > 0 {
return candidates[rand.Intn(len(candidates))], nil
sort.Slice(candidates, func(i, j int) bool { return candidates[i].LagInMB < candidates[j].LagInMB })
return spec.NamespacedName{Namespace: master.Namespace, Name: candidates[0].Name}, nil
}
if len(skipReasons) > 0 {
err = fmt.Errorf("no replica suitable for switchover: %s", strings.Join(skipReasons, `','`))
} else {
err = fmt.Errorf("no replica running")
}
return spec.NamespacedName{}, err
return spec.NamespacedName{}, fmt.Errorf("no switchover candidate found")
}
func (c *Cluster) podIsEndOfLife(pod *v1.Pod) (bool, error) {

View File

@ -38,16 +38,22 @@ func TestGetSwitchoverCandidate(t *testing.T) {
expectedError: nil,
},
{
subtest: "choose replica without a lag",
clusterJson: `{"members": [{"name": "acid-test-cluster-0", "role": "leader", "state": "running", "api_url": "http://192.168.100.1:8008/patroni", "host": "192.168.100.1", "port": 5432, "timeline": 1}, {"name": "acid-test-cluster-1", "role": "replica", "state": "running", "api_url": "http://192.168.100.2:8008/patroni", "host": "192.168.100.2", "port": 5432, "timeline": 1, "lag": 5}, {"name": "acid-test-cluster-2", "role": "replica", "state": "running", "api_url": "http://192.168.100.3:8008/patroni", "host": "192.168.100.3", "port": 5432, "timeline": 1, "lag": 0}]}`,
subtest: "choose replica with lowest lag",
clusterJson: `{"members": [{"name": "acid-test-cluster-0", "role": "leader", "state": "running", "api_url": "http://192.168.100.1:8008/patroni", "host": "192.168.100.1", "port": 5432, "timeline": 1}, {"name": "acid-test-cluster-1", "role": "replica", "state": "running", "api_url": "http://192.168.100.2:8008/patroni", "host": "192.168.100.2", "port": 5432, "timeline": 1, "lag": 5}, {"name": "acid-test-cluster-2", "role": "replica", "state": "running", "api_url": "http://192.168.100.3:8008/patroni", "host": "192.168.100.3", "port": 5432, "timeline": 1, "lag": 2}]}`,
expectedCandidate: spec.NamespacedName{Namespace: namespace, Name: "acid-test-cluster-2"},
expectedError: nil,
},
{
subtest: "no suitable replica available",
clusterJson: `{"members": [{"name": "acid-test-cluster-0", "role": "leader", "state": "running", "api_url": "http://192.168.100.1:8008/patroni", "host": "192.168.100.1", "port": 5432, "timeline": 1}, {"name": "acid-test-cluster-1", "role": "replica", "state": "running", "api_url": "http://192.168.100.2:8008/patroni", "host": "192.168.100.2", "port": 5432, "timeline": 1, "lag": 5}]}`,
subtest: "choose first replica when lag is equal evrywhere",
clusterJson: `{"members": [{"name": "acid-test-cluster-0", "role": "leader", "state": "running", "api_url": "http://192.168.100.1:8008/patroni", "host": "192.168.100.1", "port": 5432, "timeline": 1}, {"name": "acid-test-cluster-1", "role": "replica", "state": "running", "api_url": "http://192.168.100.2:8008/patroni", "host": "192.168.100.2", "port": 5432, "timeline": 1, "lag": 5}, {"name": "acid-test-cluster-2", "role": "replica", "state": "running", "api_url": "http://192.168.100.3:8008/patroni", "host": "192.168.100.3", "port": 5432, "timeline": 1, "lag": 5}]}`,
expectedCandidate: spec.NamespacedName{Namespace: namespace, Name: "acid-test-cluster-1"},
expectedError: nil,
},
{
subtest: "no running replica available",
clusterJson: `{"members": [{"name": "acid-test-cluster-0", "role": "leader", "state": "running", "api_url": "http://192.168.100.1:8008/patroni", "host": "192.168.100.1", "port": 5432, "timeline": 2}, {"name": "acid-test-cluster-1", "role": "replica", "state": "starting", "api_url": "http://192.168.100.2:8008/patroni", "host": "192.168.100.2", "port": 5432, "timeline": 2}]}`,
expectedCandidate: spec.NamespacedName{},
expectedError: fmt.Errorf("no replica suitable for switchover: acid-test-cluster-1 lags behind by 5 MB"),
expectedError: fmt.Errorf("no switchover candidate found"),
},
}