Upgrade etcd from 3.3 to 3.4

    In the general case, upgrading from etcd 3.3 to 3.4 can be a zero-downtime, rolling upgrade:

    • one by one, stop the etcd v3.3 processes and replace them with etcd v3.4 processes
    • after running all v3.4 processes, new features in v3.4 are available to the cluster

    Before , read through the rest of this guide to prepare.

    NOTE: When migrating from v2 with no v3 data, etcd server v3.2+ panics when etcd restores from existing snapshots but no v3 ETCD_DATA_DIR/member/snap/db file. This happens when the server had migrated from v2 with no previous v3 data. This also prevents accidental v3 data loss (e.g. db file might have been moved). etcd requires that post v3 migration can only happen with v3 data. Do not upgrade to newer v3 versions until v3.0 server contains v3 data.

    Highlighted breaking changes in 3.4.

    Make ETCDCTL_API=3 etcdctl default

    ETCDCTL_API=3 is now the default.

    Make etcd --enable-v2=false default

    is now the default.

    This means, unless etcd --enable-v2=true is specified, etcd v3.4 server would not serve v2 API requests.

    If v2 API were used, make sure to enable v2 API in v3.4:

    1. -etcd
    2. +etcd --enable-v2=true

    Other HTTP APIs will still work (e.g. [CLIENT-URL]/metrics, [CLIENT-URL]/health, v3 gRPC gateway).

    Deprecated etcd --ca-file and etcd --peer-ca-file flags

    --ca-file and --peer-ca-file flags are deprecated; they have been deprecated since v2.1.

    1. -etcd --ca-file ca-client.crt
    2. +etcd --trusted-ca-file ca-client.crt
    1. -etcd --peer-ca-file ca-peer.crt
    2. +etcd --peer-trusted-ca-file ca-peer.crt

    Deprecated grpc.ErrClientConnClosing error

    grpc.ErrClientConnClosing has been deprecated in gRPC >= 1.10.

    1. import (
    2. + "go.etcd.io/etcd/clientv3"
    3. "google.golang.org/grpc"
    4. + "google.golang.org/grpc/codes"
    5. + "google.golang.org/grpc/status"
    6. )
    7. _, err := kvc.Get(ctx, "a")
    8. -if err == grpc.ErrClientConnClosing {
    9. +if clientv3.IsConnCanceled(err) {
    10. // or
    11. +s, ok := status.FromError(err)
    12. +if ok {
    13. + if s.Code() == codes.Canceled

    Require grpc.WithBlock for client dial

    uses an asynchronous resolver to pass endpoints to the gRPC dial function. As a result, v3.4 client requires grpc.WithBlock dial option to wait until the underlying connection is up.

    1. import (
    2. "time"
    3. "go.etcd.io/etcd/clientv3"
    4. + "google.golang.org/grpc"
    5. )
    6. +// "grpc.WithBlock()" to block until the underlying connection is up
    7. ccfg := clientv3.Config{
    8. Endpoints: []string{"localhost:2379"},
    9. DialTimeout: time.Second,
    10. + DialOptions: []grpc.DialOption{grpc.WithBlock()},
    11. DialKeepAliveTime: time.Second,
    12. DialKeepAliveTimeout: 500 * time.Millisecond,
    13. }

    Deprecating etcd_debugging_mvcc_db_total_size_in_bytes Prometheus metrics

    v3.4 promotes etcd_debugging_mvcc_db_total_size_in_bytes Prometheus metrics to etcd_mvcc_db_total_size_in_bytes, in order to encourage etcd storage monitoring.

    etcd_debugging_mvcc_db_total_size_in_bytes is still served in v3.4 for backward compatibilities. It will be completely deprecated in v3.5.

    1. -etcd_debugging_mvcc_db_total_size_in_bytes
    2. +etcd_mvcc_db_total_size_in_bytes

    Note that etcd_debugging_* namespace metrics have been marked as experimental. As we improve monitoring guide, we may promote more metrics.

    Deprecating etcd_debugging_mvcc_put_total Prometheus metrics

    v3.4 promotes etcd_debugging_mvcc_put_total Prometheus metrics to etcd_mvcc_put_total, in order to encourage etcd storage monitoring.

    etcd_debugging_mvcc_put_total is still served in v3.4 for backward compatibilities. It will be completely deprecated in v3.5.

    1. -etcd_debugging_mvcc_put_total
    2. +etcd_mvcc_put_total

    Note that etcd_debugging_* namespace metrics have been marked as experimental. As we improve monitoring guide, we may promote more metrics.

    Deprecating etcd_debugging_mvcc_delete_total Prometheus metrics

    v3.4 promotes etcd_debugging_mvcc_delete_total Prometheus metrics to etcd_mvcc_delete_total, in order to encourage etcd storage monitoring.

    etcd_debugging_mvcc_delete_total is still served in v3.4 for backward compatibilities. It will be completely deprecated in v3.5.

    1. -etcd_debugging_mvcc_delete_total
    2. +etcd_mvcc_delete_total

    Note that etcd_debugging_* namespace metrics have been marked as experimental. As we improve monitoring guide, we may promote more metrics.

    Deprecating etcd_debugging_mvcc_txn_total Prometheus metrics

    v3.4 promotes etcd_debugging_mvcc_txn_total Prometheus metrics to etcd_mvcc_txn_total, in order to encourage etcd storage monitoring.

    etcd_debugging_mvcc_txn_total is still served in v3.4 for backward compatibilities. It will be completely deprecated in v3.5.

    1. -etcd_debugging_mvcc_txn_total
    2. +etcd_mvcc_txn_total

    Note that etcd_debugging_* namespace metrics have been marked as experimental. As we improve monitoring guide, we may promote more metrics.

    Deprecating etcd_debugging_mvcc_range_total Prometheus metrics

    v3.4 promotes etcd_debugging_mvcc_range_total Prometheus metrics to etcd_mvcc_range_total, in order to encourage etcd storage monitoring.

    Note that etcd_debugging_* namespace metrics have been marked as experimental. As we improve monitoring guide, we may promote more metrics.

    Deprecating etcd --log-output flag (now --log-outputs)

    Rename etcd --log-output to --log-outputs to support multiple log outputs. etcd --logger=capnslog does not support multiple log outputs.

    etcd --log-output will be deprecated in v3.5. etcd --logger=capnslog will be deprecated in v3.5.

    1. -etcd --log-output=stderr
    2. +etcd --log-outputs=stderr
    3. +# to write logs to stderr and a.log file at the same time
    4. +# only "--logger=zap" supports multiple writers

    v3.4 adds etcd --logger=zap --log-outputs=stderr support for structured logging and multiple log outputs. Main motivation is to promote automated etcd monitoring, rather than looking back server logs when it starts breaking. Future development will make etcd log as few as possible, and make etcd easier to monitor with metrics and alerts. etcd --logger=capnslog will be deprecated in v3.5.

    Changed log-outputs field type in etcd --config-file to []string

    Now that log-outputs (old field name log-output) accepts multiple writers, etcd configuration YAML file log-outputs field must be changed to []string type as below:

    1. # Specify 'stdout' or 'stderr' to skip journald logging even when running under systemd.
    2. -log-output: default
    3. +log-outputs: [default]

    Renamed embed.Config.LogOutput to embed.Config.LogOutputs

    Renamed to support multiple log outputs. And changed embed.Config.LogOutput type from string to []string to support multiple log outputs.

    1. import "github.com/coreos/etcd/embed"
    2. -cfg.LogOutput = "stderr"
    3. +cfg.LogOutputs = []string{"stderr"}

    v3.5 deprecates capnslog

    v3.5 will deprecate etcd --log-package-levels flag for capnslog; etcd --logger=zap --log-outputs=stderr will the default. v3.5 will deprecate [CLIENT-URL]/config/local/log endpoint.

    1. -etcd
    2. +etcd --logger zap

    Deprecating etcd --debug flag (now --log-level=debug)

    v3.4 deprecates flag. Instead, use etcd --log-level=debug flag.

    1. -etcd --debug
    2. +etcd --logger zap --log-level debug

    Deprecated pkg/transport.TLSInfo.CAFile field

    Deprecated pkg/transport.TLSInfo.CAFile field.

    1. import "github.com/coreos/etcd/pkg/transport"
    2. tlsInfo := transport.TLSInfo{
    3. CertFile: "/tmp/test-certs/test.pem",
    4. KeyFile: "/tmp/test-certs/test-key.pem",
    5. - CAFile: "/tmp/test-certs/trusted-ca.pem",
    6. + TrustedCAFile: "/tmp/test-certs/trusted-ca.pem",
    7. }
    8. tlsConfig, err := tlsInfo.ClientConfig()
    9. if err != nil {
    10. panic(err)
    11. }

    Changed embed.Config.SnapCount to embed.Config.SnapshotCount

    To be consistent with the flag name etcd --snapshot-count, embed.Config.SnapCount field has been renamed to embed.Config.SnapshotCount:

    1. import "github.com/coreos/etcd/embed"
    2. cfg := embed.NewConfig()
    3. -cfg.SnapCount = 100000
    4. +cfg.SnapshotCount = 100000

    Changed etcdserver.ServerConfig.SnapCount to etcdserver.ServerConfig.SnapshotCount

    To be consistent with the flag name etcd --snapshot-count, etcdserver.ServerConfig.SnapCount field has been renamed to etcdserver.ServerConfig.SnapshotCount:

    1. import "github.com/coreos/etcd/etcdserver"
    2. srvcfg := etcdserver.ServerConfig{
    3. - SnapCount: 100000,
    4. + SnapshotCount: 100000,

    Changed function signature in package wal

    Changed wal function signatures to support structured logger.

    1. import "github.com/coreos/etcd/wal"
    2. +import "go.uber.org/zap"
    3. +lg, _ = zap.NewProduction()
    4. -wal.Open(dirpath, snap)
    5. +wal.Open(lg, dirpath, snap)
    6. -wal.OpenForRead(dirpath, snap)
    7. +wal.OpenForRead(lg, dirpath, snap)
    8. -wal.Repair(dirpath)
    9. +wal.Repair(lg, dirpath)
    10. -wal.Create(dirpath, metadata)
    11. +wal.Create(lg, dirpath, metadata)

    Changed IntervalTree type in package pkg/adt

    pkg/adt.IntervalTree is now defined as an interface.

    Deprecated embed.Config.SetupLogging

    embed.Config.SetupLogging has been removed in order to prevent wrong logging configuration, and now set up automatically.

    1. import "github.com/coreos/etcd/embed"
    2. cfg := &embed.Config{Debug: false}
    3. -cfg.SetupLogging()

    Changed gRPC gateway HTTP endpoints (replaced /v3beta with /v3)

    Before

    1. curl -L http://localhost:2379/v3beta/kv/put \
    2. -X POST -d '{"key": "Zm9v", "value": "YmFy"}'

    After

    1. curl -L http://localhost:2379/v3/kv/put \
    2. -X POST -d '{"key": "Zm9v", "value": "YmFy"}'

    Requests to /v3beta endpoints will redirect to /v3, and /v3beta will be removed in 3.5 release.

    Deprecated container image tags

    latest and minor version images tags are deprecated:

    1. -docker pull gcr.io/etcd-development/etcd:latest
    2. +docker pull gcr.io/etcd-development/etcd:v3.4.0
    3. -docker pull gcr.io/etcd-development/etcd:v3.4
    4. +docker pull gcr.io/etcd-development/etcd:v3.4.0
    5. -docker pull gcr.io/etcd-development/etcd:v3.4
    6. +docker pull gcr.io/etcd-development/etcd:v3.4.1
    7. -docker pull gcr.io/etcd-development/etcd:v3.4
    8. +docker pull gcr.io/etcd-development/etcd:v3.4.2

    Upgrade requirements

    To upgrade an existing etcd deployment to 3.4, the running cluster must be 3.3 or greater. If it’s before 3.3, please upgrade to 3.3 before upgrading to 3.4.

    Also, to ensure a smooth rolling upgrade, the running cluster must be healthy. Check the health of the cluster by using the etcdctl endpoint health command before proceeding.

    Preparation

    Before upgrading etcd, always test the services relying on etcd in a staging environment before deploying the upgrade to the production environment.

    Before beginning, . Should something go wrong with the upgrade, it is possible to use this backup to downgrade back to existing etcd version. Please note that the snapshot command only backs up the v3 data. For v2 data, see .

    Mixed versions

    While upgrading, an etcd cluster supports mixed versions of etcd members, and operates with the protocol of the lowest common version. The cluster is only considered upgraded once all of its members are upgraded to version 3.4. Internally, etcd members negotiate with each other to determine the overall cluster version, which controls the reported version and the supported features.

    Limitations

    Note: If the cluster only has v3 data and no v2 data, it is not subject to this limitation.

    If the cluster is serving a v2 data set larger than 50MB, each newly upgraded member may take up to two minutes to catch up with the existing cluster. Check the size of a recent snapshot to estimate the total data size. In other words, it is safest to wait for 2 minutes between upgrading each member.

    Downgrade

    If all members have been upgraded to v3.4, the cluster will be upgraded to v3.4, and downgrade from this completed state is not possible. If any single member is still v3.3, however, the cluster and its operations remains “v3.3”, and it is possible from this mixed cluster state to return to using a v3.3 etcd binary on all members.

    Please download the snapshot backup to make downgrading the cluster possible even after it has been completely upgraded.

    This example shows how to upgrade a 3-member v3.3 ectd cluster running on a local machine.

    Step 1: check upgrade requirements

    Is the cluster healthy and running v3.3.x?

    1. etcdctl --endpoints=localhost:2379,localhost:22379,localhost:32379 endpoint health
    2. <<COMMENT
    3. localhost:2379 is healthy: successfully committed proposal: took = 2.118638ms
    4. localhost:22379 is healthy: successfully committed proposal: took = 3.631388ms
    5. localhost:32379 is healthy: successfully committed proposal: took = 2.157051ms
    6. COMMENT
    7. <<COMMENT
    8. {"etcdserver":"3.3.5","etcdcluster":"3.3.0"}
    9. COMMENT
    10. curl http://localhost:22379/version
    11. <<COMMENT
    12. {"etcdserver":"3.3.5","etcdcluster":"3.3.0"}
    13. COMMENT
    14. curl http://localhost:32379/version
    15. <<COMMENT
    16. {"etcdserver":"3.3.5","etcdcluster":"3.3.0"}
    17. COMMENT

    Step 2: download snapshot backup from leader

    to provide a downgrade path should any problems occur.

    etcd leader is guaranteed to have the latest application data, thus fetch snapshot from leader:

    1. curl -sL http://localhost:2379/metrics | grep etcd_server_is_leader
    2. <<COMMENT
    3. # HELP etcd_server_is_leader Whether or not this member is a leader. 1 if is, 0 otherwise.
    4. # TYPE etcd_server_is_leader gauge
    5. etcd_server_is_leader 1
    6. COMMENT
    7. curl -sL http://localhost:22379/metrics | grep etcd_server_is_leader
    8. <<COMMENT
    9. COMMENT
    10. curl -sL http://localhost:32379/metrics | grep etcd_server_is_leader
    11. <<COMMENT
    12. etcd_server_is_leader 0
    13. COMMENT
    14. etcdctl --endpoints=localhost:2379 snapshot save backup.db
    15. <<COMMENT
    16. {"level":"info","ts":1526585787.148433,"caller":"snapshot/v3_snapshot.go:109","msg":"created temporary db file","path":"backup.db.part"}
    17. {"level":"info","ts":1526585787.1485257,"caller":"snapshot/v3_snapshot.go:120","msg":"fetching snapshot","endpoint":"localhost:2379"}
    18. {"level":"info","ts":1526585787.1519694,"caller":"snapshot/v3_snapshot.go:133","msg":"fetched snapshot","endpoint":"localhost:2379","took":0.003502721}
    19. {"level":"info","ts":1526585787.1520295,"caller":"snapshot/v3_snapshot.go:142","msg":"saved","path":"backup.db"}
    20. Snapshot saved at backup.db
    21. COMMENT

    Step 3: stop one existing etcd server

    When each etcd process is stopped, expected errors will be logged by other cluster members. This is normal since a cluster member connection has been (temporarily) broken:

    1. 10.237579 I | etcdserver: updating the cluster version from 3.0 to 3.3
    2. 10.238315 N | etcdserver/membership: updated the cluster version from 3.0 to 3.3
    3. 10.238451 I | etcdserver/api: enabled capabilities for version 3.3
    4. ^C21.192174 N | pkg/osutil: received interrupt signal, shutting down...
    5. 21.192459 I | etcdserver: 7339c4e5e833c029 starts leadership transfer from 7339c4e5e833c029 to 729934363faa4a24
    6. 21.192569 I | raft: 7339c4e5e833c029 [term 8] starts to transfer leadership to 729934363faa4a24
    7. 21.192619 I | raft: 7339c4e5e833c029 sends MsgTimeoutNow to 729934363faa4a24 immediately as 729934363faa4a24 already has up-to-date log
    8. WARNING: 2018/05/17 12:45:21 grpc: addrConn.resetTransport failed to create client transport: connection error: desc = "transport: Error while dialing dial tcp: operation was canceled"; Reconnecting to {localhost:2379 0 <nil>}
    9. WARNING: 2018/05/17 12:45:21 grpc: addrConn.transportMonitor exits due to: grpc: the connection is closing
    10. 21.193589 I | raft: 7339c4e5e833c029 [term: 8] received a MsgVote message with higher term from 729934363faa4a24 [term: 9]
    11. 21.193626 I | raft: 7339c4e5e833c029 became follower at term 9
    12. 21.193651 I | raft: 7339c4e5e833c029 [logterm: 8, index: 9, vote: 0] cast MsgVote for 729934363faa4a24 [logterm: 8, index: 9] at term 9
    13. 21.193675 I | raft: raft.node: 7339c4e5e833c029 lost leader 7339c4e5e833c029 at term 9
    14. 21.194424 I | raft: raft.node: 7339c4e5e833c029 elected leader 729934363faa4a24 at term 9
    15. 21.292898 I | etcdserver: 7339c4e5e833c029 finished leadership transfer from 7339c4e5e833c029 to 729934363faa4a24 (took 100.436391ms)
    16. 21.292975 I | rafthttp: stopping peer 729934363faa4a24...
    17. 21.293206 I | rafthttp: closed the TCP streaming connection with peer 729934363faa4a24 (stream MsgApp v2 writer)
    18. 21.293225 I | rafthttp: stopped streaming with peer 729934363faa4a24 (writer)
    19. 21.293437 I | rafthttp: closed the TCP streaming connection with peer 729934363faa4a24 (stream Message writer)
    20. 21.293459 I | rafthttp: stopped streaming with peer 729934363faa4a24 (writer)
    21. 21.293514 I | rafthttp: stopped HTTP pipelining with peer 729934363faa4a24
    22. 21.293590 W | rafthttp: lost the TCP streaming connection with peer 729934363faa4a24 (stream MsgApp v2 reader)
    23. 21.293610 I | rafthttp: stopped streaming with peer 729934363faa4a24 (stream MsgApp v2 reader)
    24. 21.293680 W | rafthttp: lost the TCP streaming connection with peer 729934363faa4a24 (stream Message reader)
    25. 21.293700 I | rafthttp: stopped streaming with peer 729934363faa4a24 (stream Message reader)
    26. 21.293711 I | rafthttp: stopped peer 729934363faa4a24
    27. 21.293720 I | rafthttp: stopping peer b548c2511513015...
    28. 21.293987 I | rafthttp: closed the TCP streaming connection with peer b548c2511513015 (stream MsgApp v2 writer)
    29. 21.294063 I | rafthttp: stopped streaming with peer b548c2511513015 (writer)
    30. 21.294467 I | rafthttp: closed the TCP streaming connection with peer b548c2511513015 (stream Message writer)
    31. 21.294561 I | rafthttp: stopped streaming with peer b548c2511513015 (writer)
    32. 21.294742 I | rafthttp: stopped HTTP pipelining with peer b548c2511513015
    33. 21.294867 W | rafthttp: lost the TCP streaming connection with peer b548c2511513015 (stream MsgApp v2 reader)
    34. 21.294892 I | rafthttp: stopped streaming with peer b548c2511513015 (stream MsgApp v2 reader)
    35. 21.294990 W | rafthttp: lost the TCP streaming connection with peer b548c2511513015 (stream Message reader)
    36. 21.295004 E | rafthttp: failed to read b548c2511513015 on stream Message (context canceled)
    37. 21.295013 I | rafthttp: peer b548c2511513015 became inactive
    38. 21.295024 I | rafthttp: stopped streaming with peer b548c2511513015 (stream Message reader)
    39. 21.295035 I | rafthttp: stopped peer b548c2511513015

    Step 4: restart the etcd server with same configuration

    Restart the etcd server with same configuration but with the new etcd binary.

    1. -etcd-old --name s1 \
    2. +etcd-new --name s1 \
    3. --data-dir /tmp/etcd/s1 \
    4. --listen-client-urls http://localhost:2379 \
    5. --advertise-client-urls http://localhost:2379 \
    6. --listen-peer-urls http://localhost:2380 \
    7. --initial-advertise-peer-urls http://localhost:2380 \
    8. --initial-cluster s1=http://localhost:2380,s2=http://localhost:22380,s3=http://localhost:32380 \
    9. --initial-cluster-token tkn \
    10. + --initial-cluster-state new \
    11. + --logger zap \
    12. + --log-outputs stderr

    The new v3.4 etcd will publish its information to the cluster. At this point, cluster still operates as v3.3 protocol, which is the lowest common version.

    Verify that each member, and then the entire cluster, becomes healthy with the new v3.4 etcd binary:

    1. etcdctl endpoint health --endpoints=localhost:2379,localhost:22379,localhost:32379
    2. <<COMMENT
    3. localhost:32379 is healthy: successfully committed proposal: took = 2.337471ms
    4. localhost:22379 is healthy: successfully committed proposal: took = 1.130717ms
    5. localhost:2379 is healthy: successfully committed proposal: took = 2.124843ms
    6. COMMENT

    Un-upgraded members will log warnings like the following until the entire cluster is upgraded.

    This is expected and will cease after all etcd cluster members are upgraded to v3.4:

    Step 5: repeat step 3 and step 4 for rest of the members

    When all members are upgraded, the cluster will report upgrading to 3.4 successfully:

    Member 1:

    Member 2:

    Member 3:

    1. endpoint health --endpoints=localhost:2379,localhost:22379,localhost:32379
    2. <<COMMENT
    3. localhost:2379 is healthy: successfully committed proposal: took = 492.834µs
    4. localhost:22379 is healthy: successfully committed proposal: took = 1.015025ms
    5. localhost:32379 is healthy: successfully committed proposal: took = 1.853077ms
    6. COMMENT
    7. curl http://localhost:2379/version
    8. <<COMMENT
    9. {"etcdserver":"3.4.0","etcdcluster":"3.4.0"}
    10. COMMENT
    11. curl http://localhost:22379/version
    12. <<COMMENT
    13. {"etcdserver":"3.4.0","etcdcluster":"3.4.0"}
    14. COMMENT
    15. curl http://localhost:32379/version
    16. <<COMMENT
    17. COMMENT