From 2ab5be67a94c95fae84d2f16f54b25675b26e1a1 Mon Sep 17 00:00:00 2001 From: lengyuexuexuan <46274877+lengyuexuexuan@users.noreply.github.com> Date: Thu, 20 Jun 2024 11:24:57 +0800 Subject: [PATCH] fix(go-client): update config once replica server failed and forward to primary meta server if it was changed (#1916) https://github.com/apache/incubator-pegasus/issues/1880 https://github.com/apache/incubator-pegasus/issues/1856 As for https://github.com/apache/incubator-pegasus/issues/1856: when go client is writing to one partition and the replica node core dump, go client will finish after timeout without updating the configuration. In this case, the go client only restart to solve the problem. In this pr, the client would update configuration of table automatically when someone replica core dump. After testing, we found that the the replica error is "context.DeadlineExceeded" (incubator-pegasus/go-client/pegasus/table_connector.go) when the replica core dump. Therefore, when client meets the error, the go client will update configuration automatically. Besides, this request will not retry. Because only in the case of timeout, the configuration will be automatically updated. If you try again before then, it will still fail. There is also the risk of infinite retries. Therefore, it is better to directly return the request error to the user and let the user try again. As for https://github.com/apache/incubator-pegasus/issues/1880: When the client sends an RPC message "RPC_CM_QUERY_PARTITION_CONFIG_BY_INDEX" to the meta server, if the meta server isn't primary, the response that forward to the primary meta server will return. According to the above description, assuming that the client does not have a primary meta server configured, we can connect to the primary meta server in this way. About tests: 1. Start onebox, and the primary meta server is not added to the go client configuration. 2. The go client writes data to a certain partition and then kills the replica process. --- go-client/pegasus/table_connector.go | 1 + go-client/pegasus/table_connector_test.go | 8 ++- go-client/session/meta_call.go | 70 +++++++++++++++++++---- go-client/session/meta_session.go | 18 +++++- go-client/session/meta_session_test.go | 18 +++++- 5 files changed, 100 insertions(+), 15 deletions(-) diff --git a/go-client/pegasus/table_connector.go b/go-client/pegasus/table_connector.go index d1074a7bf3..c83461e393 100644 --- a/go-client/pegasus/table_connector.go +++ b/go-client/pegasus/table_connector.go @@ -703,6 +703,7 @@ func (p *pegasusTableConnector) handleReplicaError(err error, replica *session.R case base.ERR_TIMEOUT: case context.DeadlineExceeded: + confUpdate = true case context.Canceled: // timeout will not trigger a configuration update diff --git a/go-client/pegasus/table_connector_test.go b/go-client/pegasus/table_connector_test.go index 1b28747655..b4016748ea 100644 --- a/go-client/pegasus/table_connector_test.go +++ b/go-client/pegasus/table_connector_test.go @@ -269,8 +269,14 @@ func TestPegasusTableConnector_TriggerSelfUpdate(t *testing.T) { assert.True(t, confUpdate) assert.False(t, retry) + confUpdate, retry, err = ptb.handleReplicaError(context.DeadlineExceeded, nil) + <-ptb.confUpdateCh + assert.Error(t, err) + assert.True(t, confUpdate) + assert.False(t, retry) + { // Ensure: The following errors should not trigger configuration update - errorTypes := []error{base.ERR_TIMEOUT, context.DeadlineExceeded, base.ERR_CAPACITY_EXCEEDED, base.ERR_NOT_ENOUGH_MEMBER, base.ERR_BUSY, base.ERR_SPLITTING, base.ERR_DISK_INSUFFICIENT} + errorTypes := []error{base.ERR_TIMEOUT, base.ERR_CAPACITY_EXCEEDED, base.ERR_NOT_ENOUGH_MEMBER, base.ERR_BUSY, base.ERR_SPLITTING, base.ERR_DISK_INSUFFICIENT} for _, err := range errorTypes { channelEmpty := false diff --git a/go-client/session/meta_call.go b/go-client/session/meta_call.go index 2db6179ab1..d846aa09b8 100644 --- a/go-client/session/meta_call.go +++ b/go-client/session/meta_call.go @@ -26,6 +26,8 @@ import ( "time" "github.com/apache/incubator-pegasus/go-client/idl/base" + "github.com/apache/incubator-pegasus/go-client/idl/replication" + "github.com/apache/incubator-pegasus/go-client/pegalog" ) type metaCallFunc func(context.Context, *metaSession) (metaResponse, error) @@ -42,21 +44,24 @@ type metaCall struct { backupCh chan interface{} callFunc metaCallFunc - metas []*metaSession - lead int + metaIPAddrs []string + metas []*metaSession + lead int // After a Run successfully ends, the current leader will be set in this field. // If there is no meta failover, `newLead` equals to `lead`. newLead uint32 + lock sync.RWMutex } -func newMetaCall(lead int, metas []*metaSession, callFunc metaCallFunc) *metaCall { +func newMetaCall(lead int, metas []*metaSession, callFunc metaCallFunc, meatIPAddr []string) *metaCall { return &metaCall{ - metas: metas, - lead: lead, - newLead: uint32(lead), - respCh: make(chan metaResponse), - callFunc: callFunc, - backupCh: make(chan interface{}), + metas: metas, + metaIPAddrs: meatIPAddr, + lead: lead, + newLead: uint32(lead), + respCh: make(chan metaResponse), + callFunc: callFunc, + backupCh: make(chan interface{}), } } @@ -106,14 +111,44 @@ func (c *metaCall) Run(ctx context.Context) (metaResponse, error) { } // issueSingleMeta returns false if we should try another meta -func (c *metaCall) issueSingleMeta(ctx context.Context, i int) bool { - meta := c.metas[i] +func (c *metaCall) issueSingleMeta(ctx context.Context, curLeader int) bool { + meta := c.metas[curLeader] resp, err := c.callFunc(ctx, meta) + + if err == nil && resp.GetErr().Errno == base.ERR_FORWARD_TO_OTHERS.String() { + forwardAddr := c.getMetaServiceForwardAddress(resp) + if forwardAddr == nil { + return false + } + addr := forwardAddr.GetAddress() + found := false + c.lock.Lock() + for i := range c.metaIPAddrs { + if addr == c.metaIPAddrs[i] { + found = true + break + } + } + c.lock.Unlock() + if !found { + c.lock.Lock() + c.metaIPAddrs = append(c.metaIPAddrs, addr) + c.metas = append(c.metas, &metaSession{ + NodeSession: newNodeSession(addr, NodeTypeMeta), + logger: pegalog.GetLogger(), + }) + c.lock.Unlock() + curLeader = len(c.metas) - 1 + c.metas[curLeader].logger.Printf("add forward address %s as meta server", addr) + resp, err = c.callFunc(ctx, c.metas[curLeader]) + } + } + if err != nil || resp.GetErr().Errno == base.ERR_FORWARD_TO_OTHERS.String() { return false } // the RPC succeeds, this meta becomes the new leader now. - atomic.StoreUint32(&c.newLead, uint32(i)) + atomic.StoreUint32(&c.newLead, uint32(curLeader)) select { case <-ctx.Done(): case c.respCh <- resp: @@ -133,3 +168,14 @@ func (c *metaCall) issueBackupMetas(ctx context.Context) { }(i) } } + +func (c *metaCall) getMetaServiceForwardAddress(resp metaResponse) *base.RPCAddress { + rep, ok := resp.(*replication.QueryCfgResponse) + if !ok || rep.GetErr().Errno != base.ERR_FORWARD_TO_OTHERS.String() { + return nil + } else if rep.GetPartitions() == nil || len(rep.GetPartitions()) == 0 { + return nil + } else { + return rep.Partitions[0].Primary + } +} diff --git a/go-client/session/meta_session.go b/go-client/session/meta_session.go index c209cb8488..b0e962d1d9 100644 --- a/go-client/session/meta_session.go +++ b/go-client/session/meta_session.go @@ -94,10 +94,12 @@ func NewMetaManager(addrs []string, creator NodeSessionCreator) *MetaManager { func (m *MetaManager) call(ctx context.Context, callFunc metaCallFunc) (metaResponse, error) { lead := m.getCurrentLeader() - call := newMetaCall(lead, m.metas, callFunc) + call := newMetaCall(lead, m.metas, callFunc, m.metaIPAddrs) resp, err := call.Run(ctx) if err == nil { m.setCurrentLeader(int(call.newLead)) + m.setNewMetas(call.metas) + m.setMetaIPAddrs(call.metaIPAddrs) } return resp, err } @@ -131,6 +133,20 @@ func (m *MetaManager) setCurrentLeader(lead int) { m.currentLeader = lead } +func (m *MetaManager) setNewMetas(metas []*metaSession) { + m.mu.Lock() + defer m.mu.Unlock() + + m.metas = metas +} + +func (m *MetaManager) setMetaIPAddrs(metaIPAddrs []string) { + m.mu.Lock() + defer m.mu.Unlock() + + m.metaIPAddrs = metaIPAddrs +} + // Close the sessions. func (m *MetaManager) Close() error { funcs := make([]func() error, len(m.metas)) diff --git a/go-client/session/meta_session_test.go b/go-client/session/meta_session_test.go index d2cbf6cc3d..5014a4680e 100644 --- a/go-client/session/meta_session_test.go +++ b/go-client/session/meta_session_test.go @@ -118,7 +118,7 @@ func TestMetaManager_FirstMetaDead(t *testing.T) { for i := 0; i < 3; i++ { call := newMetaCall(mm.currentLeader, mm.metas, func(rpcCtx context.Context, ms *metaSession) (metaResponse, error) { return ms.queryConfig(rpcCtx, "temp") - }) + }, []string{"0.0.0.0:12345", "0.0.0.0:34603", "0.0.0.0:34602", "0.0.0.0:34601"}) // This a trick for testing. If metaCall issue to other meta, not only to the leader, this nil channel will cause panic. call.backupCh = nil metaResp, err := call.Run(context.Background()) @@ -126,3 +126,19 @@ func TestMetaManager_FirstMetaDead(t *testing.T) { assert.Equal(t, metaResp.GetErr().Errno, base.ERR_OK.String()) } } + +// This case mocks the case that the server primary meta is not in the client metalist. +// And the client will forward to the primary meta automatically. +func TestNodeSession_ForwardToPrimaryMeta(t *testing.T) { + defer leaktest.Check(t)() + + metaList := []string{"0.0.0.0:34601", "0.0.0.0:34602", "0.0.0.0:34603"} + + for i := 0; i < 3; i++ { + mm := NewMetaManager(metaList[i:i+1], NewNodeSession) + defer mm.Close() + resp, err := mm.QueryConfig(context.Background(), "temp") + assert.Nil(t, err) + assert.Equal(t, resp.Err.Errno, base.ERR_OK.String()) + } +}