7494a0c4fd
Nomad inherited protocol version numbering configuration from Consul and
Serf, but unlike those projects Nomad has never used it. Nomad's
`protocol_version` has always been `1`.
While the code is effectively unused and therefore poses no runtime
risks to leave, I felt like removing it was best because:
1. Nomad's RPC subsystem has been able to evolve extensively without
needing to increment the version number.
2. Nomad's HTTP API has evolved extensively without increment
`API{Major,Minor}Version`. If we want to version the HTTP API in the
future, I doubt this is the mechanism we would choose.
3. The presence of the `server.protocol_version` configuration
parameter is confusing since `server.raft_protocol` *is* an important
parameter for operators to consider. Even more confusing is that
there is a distinct Serf protocol version which is included in `nomad
server members` output under the heading `Protocol`. `raft_protocol`
is the *only* protocol version relevant to Nomad developers and
operators. The other protocol versions are either deadcode or have
never changed (Serf).
4. If we were to need to version the RPC, HTTP API, or Serf protocols, I
don't think these configuration parameters and variables are the best
choice. If we come to that point we should choose a versioning scheme
based on the use case and modern best practices -- not this 6+ year
old dead code.
103 lines
3.1 KiB
Go
103 lines
3.1 KiB
Go
package nomad
|
|
|
|
import (
|
|
"context"
|
|
"sync"
|
|
|
|
log "github.com/hashicorp/go-hclog"
|
|
|
|
"github.com/hashicorp/consul/agent/consul/autopilot"
|
|
"github.com/hashicorp/nomad/helper/pool"
|
|
"github.com/hashicorp/serf/serf"
|
|
)
|
|
|
|
// StatsFetcher has two functions for autopilot. First, lets us fetch all the
|
|
// stats in parallel so we are taking a sample as close to the same time as
|
|
// possible, since we are comparing time-sensitive info for the health check.
|
|
// Second, it bounds the time so that one slow RPC can't hold up the health
|
|
// check loop; as a side effect of how it implements this, it also limits to
|
|
// a single in-flight RPC to any given server, so goroutines don't accumulate
|
|
// as we run the health check fairly frequently.
|
|
type StatsFetcher struct {
|
|
logger log.Logger
|
|
pool *pool.ConnPool
|
|
region string
|
|
inflight map[string]struct{}
|
|
inflightLock sync.Mutex
|
|
}
|
|
|
|
// NewStatsFetcher returns a stats fetcher.
|
|
func NewStatsFetcher(logger log.Logger, pool *pool.ConnPool, region string) *StatsFetcher {
|
|
return &StatsFetcher{
|
|
logger: logger.Named("stats_fetcher"),
|
|
pool: pool,
|
|
region: region,
|
|
inflight: make(map[string]struct{}),
|
|
}
|
|
}
|
|
|
|
// fetch does the RPC to fetch the server stats from a single server. We don't
|
|
// cancel this when the context is canceled because we only want one in-flight
|
|
// RPC to each server, so we let it finish and then clean up the in-flight
|
|
// tracking.
|
|
func (f *StatsFetcher) fetch(server *serverParts, replyCh chan *autopilot.ServerStats) {
|
|
var args struct{}
|
|
var reply autopilot.ServerStats
|
|
err := f.pool.RPC(f.region, server.Addr, "Status.RaftStats", &args, &reply)
|
|
if err != nil {
|
|
f.logger.Warn("failed retrieving server health", "server", server.Name, "error", err)
|
|
} else {
|
|
replyCh <- &reply
|
|
}
|
|
|
|
f.inflightLock.Lock()
|
|
delete(f.inflight, server.ID)
|
|
f.inflightLock.Unlock()
|
|
}
|
|
|
|
// Fetch will attempt to query all the servers in parallel.
|
|
func (f *StatsFetcher) Fetch(ctx context.Context, members []serf.Member) map[string]*autopilot.ServerStats {
|
|
type workItem struct {
|
|
server *serverParts
|
|
replyCh chan *autopilot.ServerStats
|
|
}
|
|
var servers []*serverParts
|
|
for _, s := range members {
|
|
if ok, parts := isNomadServer(s); ok {
|
|
servers = append(servers, parts)
|
|
}
|
|
}
|
|
|
|
// Skip any servers that have inflight requests.
|
|
var work []*workItem
|
|
f.inflightLock.Lock()
|
|
for _, server := range servers {
|
|
if _, ok := f.inflight[server.ID]; ok {
|
|
f.logger.Warn("failed retrieving server health; last request still outstanding", "server", server.Name)
|
|
} else {
|
|
workItem := &workItem{
|
|
server: server,
|
|
replyCh: make(chan *autopilot.ServerStats, 1),
|
|
}
|
|
work = append(work, workItem)
|
|
f.inflight[server.ID] = struct{}{}
|
|
go f.fetch(workItem.server, workItem.replyCh)
|
|
}
|
|
}
|
|
f.inflightLock.Unlock()
|
|
|
|
// Now wait for the results to come in, or for the context to be
|
|
// canceled.
|
|
replies := make(map[string]*autopilot.ServerStats)
|
|
for _, workItem := range work {
|
|
select {
|
|
case reply := <-workItem.replyCh:
|
|
replies[workItem.server.ID] = reply
|
|
|
|
case <-ctx.Done():
|
|
f.logger.Warn("failed retrieving server health", "server", workItem.server.Name, "error", ctx.Err())
|
|
}
|
|
}
|
|
return replies
|
|
}
|