ipfs-cluster/monitor/pubsubmon/pubsubmon.go

246 lines
5.5 KiB
Go
Raw Normal View History

// Package pubsubmon implements a PeerMonitor component for IPFS Cluster that
// uses PubSub to send and receive metrics.
package pubsubmon
import (
"bytes"
"context"
"sync"
"github.com/ipfs/ipfs-cluster/api"
"github.com/ipfs/ipfs-cluster/monitor/metrics"
"go.opencensus.io/trace"
logging "github.com/ipfs/go-log"
rpc "github.com/libp2p/go-libp2p-gorpc"
peer "github.com/libp2p/go-libp2p-peer"
pubsub "github.com/libp2p/go-libp2p-pubsub"
msgpack "github.com/multiformats/go-multicodec/msgpack"
)
var logger = logging.Logger("monitor")
// PubsubTopic specifies the topic used to publish Cluster metrics.
var PubsubTopic = "monitor.metrics"
var msgpackHandle = msgpack.DefaultMsgpackHandle()
// Monitor is a component in charge of monitoring peers, logging
// metrics and detecting failures
type Monitor struct {
ctx context.Context
cancel func()
rpcClient *rpc.Client
rpcReady chan struct{}
pubsub *pubsub.PubSub
subscription *pubsub.Subscription
Consensus: add new "crdt" consensus component This adds a new "crdt" consensus component using go-ds-crdt. This implies several refactors to fully make cluster consensus-component independent: * Delete mapstate and fully adopt dsstate (after people have migrated). * Return errors from state methods rather than ignoring them. * Add a new "datastore" modules so that we can configure datastores in the main configuration like other components. * Let the consensus components fully define the "state.State". Thus, they do not receive the state, they receive the storage where we put the state (a go-datastore). * Allow to customize how the monitor component obtains Peers() (the current peerset), including avoiding using the current peerset. At the moment the crdt consensus uses the monitoring component to define the current peerset. Therefore the monitor component cannot rely on the consensus component to produce a peerset. * Re-factor/re-implementation of "ipfs-cluster-service state" operations. Includes the dissapearance of the "migrate" one. The CRDT consensus component defines creates a crdt-datastore (with ipfs-lite) and uses it to intitialize a dssate. Thus the crdt-store is elegantly wrapped. Any modifications to the state get automatically replicated to other peers. We store all the CRDT DAG blocks in the local datastore. The consensus components only expose a ReadOnly state, as any modifications to the shared state should happen through them. DHT and PubSub facilities must now be created outside of Cluster and passed in so they can be re-used by different components.
2019-02-20 14:24:25 +00:00
peers PeersFunc
metrics *metrics.Store
checker *metrics.Checker
config *Config
shutdownLock sync.Mutex
shutdown bool
wg sync.WaitGroup
}
Consensus: add new "crdt" consensus component This adds a new "crdt" consensus component using go-ds-crdt. This implies several refactors to fully make cluster consensus-component independent: * Delete mapstate and fully adopt dsstate (after people have migrated). * Return errors from state methods rather than ignoring them. * Add a new "datastore" modules so that we can configure datastores in the main configuration like other components. * Let the consensus components fully define the "state.State". Thus, they do not receive the state, they receive the storage where we put the state (a go-datastore). * Allow to customize how the monitor component obtains Peers() (the current peerset), including avoiding using the current peerset. At the moment the crdt consensus uses the monitoring component to define the current peerset. Therefore the monitor component cannot rely on the consensus component to produce a peerset. * Re-factor/re-implementation of "ipfs-cluster-service state" operations. Includes the dissapearance of the "migrate" one. The CRDT consensus component defines creates a crdt-datastore (with ipfs-lite) and uses it to intitialize a dssate. Thus the crdt-store is elegantly wrapped. Any modifications to the state get automatically replicated to other peers. We store all the CRDT DAG blocks in the local datastore. The consensus components only expose a ReadOnly state, as any modifications to the shared state should happen through them. DHT and PubSub facilities must now be created outside of Cluster and passed in so they can be re-used by different components.
2019-02-20 14:24:25 +00:00
// PeersFunc allows the Monitor to filter and discard metrics
// that do not belong to a given peerset.
type PeersFunc func(context.Context) ([]peer.ID, error)
// New creates a new PubSub monitor, using the given host, config and
// PeersFunc. The PeersFunc can be nil. In this case, no metric filtering is
// done based on peers (any peer is considered part of the peerset).
func New(
cfg *Config,
psub *pubsub.PubSub,
peers PeersFunc,
) (*Monitor, error) {
err := cfg.Validate()
if err != nil {
return nil, err
}
ctx, cancel := context.WithCancel(context.Background())
mtrs := metrics.NewStore()
checker := metrics.NewChecker(mtrs)
Consensus: add new "crdt" consensus component This adds a new "crdt" consensus component using go-ds-crdt. This implies several refactors to fully make cluster consensus-component independent: * Delete mapstate and fully adopt dsstate (after people have migrated). * Return errors from state methods rather than ignoring them. * Add a new "datastore" modules so that we can configure datastores in the main configuration like other components. * Let the consensus components fully define the "state.State". Thus, they do not receive the state, they receive the storage where we put the state (a go-datastore). * Allow to customize how the monitor component obtains Peers() (the current peerset), including avoiding using the current peerset. At the moment the crdt consensus uses the monitoring component to define the current peerset. Therefore the monitor component cannot rely on the consensus component to produce a peerset. * Re-factor/re-implementation of "ipfs-cluster-service state" operations. Includes the dissapearance of the "migrate" one. The CRDT consensus component defines creates a crdt-datastore (with ipfs-lite) and uses it to intitialize a dssate. Thus the crdt-store is elegantly wrapped. Any modifications to the state get automatically replicated to other peers. We store all the CRDT DAG blocks in the local datastore. The consensus components only expose a ReadOnly state, as any modifications to the shared state should happen through them. DHT and PubSub facilities must now be created outside of Cluster and passed in so they can be re-used by different components.
2019-02-20 14:24:25 +00:00
subscription, err := psub.Subscribe(PubsubTopic)
if err != nil {
cancel()
return nil, err
}
mon := &Monitor{
ctx: ctx,
cancel: cancel,
rpcReady: make(chan struct{}, 1),
Consensus: add new "crdt" consensus component This adds a new "crdt" consensus component using go-ds-crdt. This implies several refactors to fully make cluster consensus-component independent: * Delete mapstate and fully adopt dsstate (after people have migrated). * Return errors from state methods rather than ignoring them. * Add a new "datastore" modules so that we can configure datastores in the main configuration like other components. * Let the consensus components fully define the "state.State". Thus, they do not receive the state, they receive the storage where we put the state (a go-datastore). * Allow to customize how the monitor component obtains Peers() (the current peerset), including avoiding using the current peerset. At the moment the crdt consensus uses the monitoring component to define the current peerset. Therefore the monitor component cannot rely on the consensus component to produce a peerset. * Re-factor/re-implementation of "ipfs-cluster-service state" operations. Includes the dissapearance of the "migrate" one. The CRDT consensus component defines creates a crdt-datastore (with ipfs-lite) and uses it to intitialize a dssate. Thus the crdt-store is elegantly wrapped. Any modifications to the state get automatically replicated to other peers. We store all the CRDT DAG blocks in the local datastore. The consensus components only expose a ReadOnly state, as any modifications to the shared state should happen through them. DHT and PubSub facilities must now be created outside of Cluster and passed in so they can be re-used by different components.
2019-02-20 14:24:25 +00:00
pubsub: psub,
subscription: subscription,
Consensus: add new "crdt" consensus component This adds a new "crdt" consensus component using go-ds-crdt. This implies several refactors to fully make cluster consensus-component independent: * Delete mapstate and fully adopt dsstate (after people have migrated). * Return errors from state methods rather than ignoring them. * Add a new "datastore" modules so that we can configure datastores in the main configuration like other components. * Let the consensus components fully define the "state.State". Thus, they do not receive the state, they receive the storage where we put the state (a go-datastore). * Allow to customize how the monitor component obtains Peers() (the current peerset), including avoiding using the current peerset. At the moment the crdt consensus uses the monitoring component to define the current peerset. Therefore the monitor component cannot rely on the consensus component to produce a peerset. * Re-factor/re-implementation of "ipfs-cluster-service state" operations. Includes the dissapearance of the "migrate" one. The CRDT consensus component defines creates a crdt-datastore (with ipfs-lite) and uses it to intitialize a dssate. Thus the crdt-store is elegantly wrapped. Any modifications to the state get automatically replicated to other peers. We store all the CRDT DAG blocks in the local datastore. The consensus components only expose a ReadOnly state, as any modifications to the shared state should happen through them. DHT and PubSub facilities must now be created outside of Cluster and passed in so they can be re-used by different components.
2019-02-20 14:24:25 +00:00
peers: peers,
metrics: mtrs,
checker: checker,
config: cfg,
}
go mon.run()
return mon, nil
}
func (mon *Monitor) run() {
select {
case <-mon.rpcReady:
go mon.logFromPubsub()
Consensus: add new "crdt" consensus component This adds a new "crdt" consensus component using go-ds-crdt. This implies several refactors to fully make cluster consensus-component independent: * Delete mapstate and fully adopt dsstate (after people have migrated). * Return errors from state methods rather than ignoring them. * Add a new "datastore" modules so that we can configure datastores in the main configuration like other components. * Let the consensus components fully define the "state.State". Thus, they do not receive the state, they receive the storage where we put the state (a go-datastore). * Allow to customize how the monitor component obtains Peers() (the current peerset), including avoiding using the current peerset. At the moment the crdt consensus uses the monitoring component to define the current peerset. Therefore the monitor component cannot rely on the consensus component to produce a peerset. * Re-factor/re-implementation of "ipfs-cluster-service state" operations. Includes the dissapearance of the "migrate" one. The CRDT consensus component defines creates a crdt-datastore (with ipfs-lite) and uses it to intitialize a dssate. Thus the crdt-store is elegantly wrapped. Any modifications to the state get automatically replicated to other peers. We store all the CRDT DAG blocks in the local datastore. The consensus components only expose a ReadOnly state, as any modifications to the shared state should happen through them. DHT and PubSub facilities must now be created outside of Cluster and passed in so they can be re-used by different components.
2019-02-20 14:24:25 +00:00
go mon.checker.Watch(mon.ctx, mon.peers, mon.config.CheckInterval)
case <-mon.ctx.Done():
}
}
// logFromPubsub logs metrics received in the subscribed topic.
func (mon *Monitor) logFromPubsub() {
ctx, span := trace.StartSpan(mon.ctx, "monitor/pubsub/logFromPubsub")
defer span.End()
for {
select {
case <-ctx.Done():
return
default:
msg, err := mon.subscription.Next(ctx)
if err != nil { // context cancelled enters here
continue
}
data := msg.GetData()
buf := bytes.NewBuffer(data)
dec := msgpack.Multicodec(msgpackHandle).Decoder(buf)
metric := api.Metric{}
err = dec.Decode(&metric)
if err != nil {
logger.Error(err)
continue
}
logger.Debugf(
"received pubsub metric '%s' from '%s'",
metric.Name,
metric.Peer,
)
err = mon.LogMetric(ctx, &metric)
if err != nil {
logger.Error(err)
continue
}
}
}
}
// SetClient saves the given rpc.Client for later use
func (mon *Monitor) SetClient(c *rpc.Client) {
mon.rpcClient = c
mon.rpcReady <- struct{}{}
}
// Shutdown stops the peer monitor. It particular, it will
// not deliver any alerts.
func (mon *Monitor) Shutdown(ctx context.Context) error {
ctx, span := trace.StartSpan(ctx, "monitor/pubsub/Shutdown")
defer span.End()
mon.shutdownLock.Lock()
defer mon.shutdownLock.Unlock()
if mon.shutdown {
logger.Warning("Monitor already shut down")
return nil
}
logger.Info("stopping Monitor")
close(mon.rpcReady)
mon.cancel()
mon.wg.Wait()
mon.shutdown = true
return nil
}
// LogMetric stores a metric so it can later be retrieved.
func (mon *Monitor) LogMetric(ctx context.Context, m *api.Metric) error {
ctx, span := trace.StartSpan(ctx, "monitor/pubsub/LogMetric")
defer span.End()
mon.metrics.Add(m)
logger.Debugf("pubsub mon logged '%s' metric from '%s'. Expires on %d", m.Name, m.Peer, m.Expire)
return nil
}
// PublishMetric broadcasts a metric to all current cluster peers.
func (mon *Monitor) PublishMetric(ctx context.Context, m *api.Metric) error {
ctx, span := trace.StartSpan(ctx, "monitor/pubsub/PublishMetric")
defer span.End()
if m.Discard() {
logger.Warningf("discarding invalid metric: %+v", m)
return nil
}
var b bytes.Buffer
enc := msgpack.Multicodec(msgpackHandle).Encoder(&b)
err := enc.Encode(m)
if err != nil {
logger.Error(err)
return err
}
logger.Debugf(
"publishing metric %s to pubsub. Expires: %d",
m.Name,
m.Expire,
)
err = mon.pubsub.Publish(PubsubTopic, b.Bytes())
if err != nil {
logger.Error(err)
return err
}
return nil
}
// LatestMetrics returns last known VALID metrics of a given type. A metric
Consensus: add new "crdt" consensus component This adds a new "crdt" consensus component using go-ds-crdt. This implies several refactors to fully make cluster consensus-component independent: * Delete mapstate and fully adopt dsstate (after people have migrated). * Return errors from state methods rather than ignoring them. * Add a new "datastore" modules so that we can configure datastores in the main configuration like other components. * Let the consensus components fully define the "state.State". Thus, they do not receive the state, they receive the storage where we put the state (a go-datastore). * Allow to customize how the monitor component obtains Peers() (the current peerset), including avoiding using the current peerset. At the moment the crdt consensus uses the monitoring component to define the current peerset. Therefore the monitor component cannot rely on the consensus component to produce a peerset. * Re-factor/re-implementation of "ipfs-cluster-service state" operations. Includes the dissapearance of the "migrate" one. The CRDT consensus component defines creates a crdt-datastore (with ipfs-lite) and uses it to intitialize a dssate. Thus the crdt-store is elegantly wrapped. Any modifications to the state get automatically replicated to other peers. We store all the CRDT DAG blocks in the local datastore. The consensus components only expose a ReadOnly state, as any modifications to the shared state should happen through them. DHT and PubSub facilities must now be created outside of Cluster and passed in so they can be re-used by different components.
2019-02-20 14:24:25 +00:00
// is only valid if it has not expired and belongs to a current cluster peer.
func (mon *Monitor) LatestMetrics(ctx context.Context, name string) []*api.Metric {
ctx, span := trace.StartSpan(ctx, "monitor/pubsub/LatestMetrics")
defer span.End()
Consensus: add new "crdt" consensus component This adds a new "crdt" consensus component using go-ds-crdt. This implies several refactors to fully make cluster consensus-component independent: * Delete mapstate and fully adopt dsstate (after people have migrated). * Return errors from state methods rather than ignoring them. * Add a new "datastore" modules so that we can configure datastores in the main configuration like other components. * Let the consensus components fully define the "state.State". Thus, they do not receive the state, they receive the storage where we put the state (a go-datastore). * Allow to customize how the monitor component obtains Peers() (the current peerset), including avoiding using the current peerset. At the moment the crdt consensus uses the monitoring component to define the current peerset. Therefore the monitor component cannot rely on the consensus component to produce a peerset. * Re-factor/re-implementation of "ipfs-cluster-service state" operations. Includes the dissapearance of the "migrate" one. The CRDT consensus component defines creates a crdt-datastore (with ipfs-lite) and uses it to intitialize a dssate. Thus the crdt-store is elegantly wrapped. Any modifications to the state get automatically replicated to other peers. We store all the CRDT DAG blocks in the local datastore. The consensus components only expose a ReadOnly state, as any modifications to the shared state should happen through them. DHT and PubSub facilities must now be created outside of Cluster and passed in so they can be re-used by different components.
2019-02-20 14:24:25 +00:00
latest := mon.metrics.LatestValid(name)
if mon.peers == nil {
return latest
}
Consensus: add new "crdt" consensus component This adds a new "crdt" consensus component using go-ds-crdt. This implies several refactors to fully make cluster consensus-component independent: * Delete mapstate and fully adopt dsstate (after people have migrated). * Return errors from state methods rather than ignoring them. * Add a new "datastore" modules so that we can configure datastores in the main configuration like other components. * Let the consensus components fully define the "state.State". Thus, they do not receive the state, they receive the storage where we put the state (a go-datastore). * Allow to customize how the monitor component obtains Peers() (the current peerset), including avoiding using the current peerset. At the moment the crdt consensus uses the monitoring component to define the current peerset. Therefore the monitor component cannot rely on the consensus component to produce a peerset. * Re-factor/re-implementation of "ipfs-cluster-service state" operations. Includes the dissapearance of the "migrate" one. The CRDT consensus component defines creates a crdt-datastore (with ipfs-lite) and uses it to intitialize a dssate. Thus the crdt-store is elegantly wrapped. Any modifications to the state get automatically replicated to other peers. We store all the CRDT DAG blocks in the local datastore. The consensus components only expose a ReadOnly state, as any modifications to the shared state should happen through them. DHT and PubSub facilities must now be created outside of Cluster and passed in so they can be re-used by different components.
2019-02-20 14:24:25 +00:00
// Make sure we only return metrics in the current peerset if we have
// a peerset provider.
peers, err := mon.peers(ctx)
if err != nil {
return []*api.Metric{}
}
return metrics.PeersetFilter(latest, peers)
}
// Alerts returns a channel on which alerts are sent when the
// monitor detects a failure.
func (mon *Monitor) Alerts() <-chan *api.Alert {
return mon.checker.Alerts()
}