syncthing/cmd/stdiscosrv/database.go

349 lines
8.9 KiB
Go
Raw Normal View History

cmd/stdiscosrv: New discovery server (fixes #4618) This is a new revision of the discovery server. Relevant changes and non-changes: - Protocol towards clients is unchanged. - Recommended large scale design is still to be deployed nehind nginx (I tested, and it's still a lot faster at terminating TLS). - Database backend is leveldb again, only. It scales enough, is easy to setup, and we don't need any backend to take care of. - Server supports replication. This is a simple TCP channel - protect it with a firewall when deploying over the internet. (We deploy this within the same datacenter, and with firewall.) Any incoming client announces are sent over the replication channel(s) to other peer discosrvs. Incoming replication changes are applied to the database as if they came from clients, but without the TLS/certificate overhead. - Metrics are exposed using the prometheus library, when enabled. - The database values and replication protocol is protobuf, because JSON was quite CPU intensive when I tried that and benchmarked it. - The "Retry-After" value for failed lookups gets slowly increased from a default of 120 seconds, by 5 seconds for each failed lookup, independently by each discosrv. This lowers the query load over time for clients that are never seen. The Retry-After maxes out at 3600 after a couple of weeks of this increase. The number of failed lookups is stored in the database, now and then (avoiding making each lookup a database put). All in all this means clients can be pointed towards a cluster using just multiple A / AAAA records to gain both load sharing and redundancy (if one is down, clients will talk to the remaining ones). GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4648
2018-01-14 08:52:31 +00:00
// Copyright (C) 2018 The Syncthing Authors.
//
// This Source Code Form is subject to the terms of the Mozilla Public
// License, v. 2.0. If a copy of the MPL was not distributed with this file,
// You can obtain one at https://mozilla.org/MPL/2.0/.
//go:generate go run ../../script/protofmt.go database.proto
//go:generate protoc -I ../../../../../ -I ../../vendor/ -I ../../vendor/github.com/gogo/protobuf/protobuf -I . --gogofast_out=. database.proto
package main
import (
"sort"
"time"
"github.com/syndtr/goleveldb/leveldb"
"github.com/syndtr/goleveldb/leveldb/util"
)
type clock interface {
Now() time.Time
}
type defaultClock struct{}
func (defaultClock) Now() time.Time {
return time.Now()
}
type database interface {
put(key string, rec DatabaseRecord) error
merge(key string, addrs []DatabaseAddress, seen int64) error
get(key string) (DatabaseRecord, error)
}
type levelDBStore struct {
db *leveldb.DB
inbox chan func()
stop chan struct{}
clock clock
marshalBuf []byte
}
func newLevelDBStore(dir string) (*levelDBStore, error) {
db, err := leveldb.OpenFile(dir, levelDBOptions)
if err != nil {
return nil, err
}
return &levelDBStore{
db: db,
inbox: make(chan func(), 16),
stop: make(chan struct{}),
clock: defaultClock{},
}, nil
}
func (s *levelDBStore) put(key string, rec DatabaseRecord) error {
t0 := time.Now()
defer func() {
databaseOperationSeconds.WithLabelValues(dbOpPut).Observe(time.Since(t0).Seconds())
}()
rc := make(chan error)
s.inbox <- func() {
size := rec.Size()
if len(s.marshalBuf) < size {
s.marshalBuf = make([]byte, size)
}
n, _ := rec.MarshalTo(s.marshalBuf)
rc <- s.db.Put([]byte(key), s.marshalBuf[:n], nil)
}
err := <-rc
if err != nil {
databaseOperations.WithLabelValues(dbOpPut, dbResError).Inc()
} else {
databaseOperations.WithLabelValues(dbOpPut, dbResSuccess).Inc()
}
return err
}
func (s *levelDBStore) merge(key string, addrs []DatabaseAddress, seen int64) error {
t0 := time.Now()
defer func() {
databaseOperationSeconds.WithLabelValues(dbOpMerge).Observe(time.Since(t0).Seconds())
}()
rc := make(chan error)
newRec := DatabaseRecord{
Addresses: addrs,
Seen: seen,
}
s.inbox <- func() {
// grab the existing record
oldRec, err := s.get(key)
if err != nil {
// "not found" is not an error from get, so this is serious
// stuff only
rc <- err
return
}
newRec = merge(newRec, oldRec)
// We replicate s.put() functionality here ourselves instead of
// calling it because we want to serialize our get above together
// with the put in the same function.
size := newRec.Size()
if len(s.marshalBuf) < size {
s.marshalBuf = make([]byte, size)
}
n, _ := newRec.MarshalTo(s.marshalBuf)
rc <- s.db.Put([]byte(key), s.marshalBuf[:n], nil)
}
err := <-rc
if err != nil {
databaseOperations.WithLabelValues(dbOpMerge, dbResError).Inc()
} else {
databaseOperations.WithLabelValues(dbOpMerge, dbResSuccess).Inc()
}
return err
}
func (s *levelDBStore) get(key string) (DatabaseRecord, error) {
t0 := time.Now()
defer func() {
databaseOperationSeconds.WithLabelValues(dbOpGet).Observe(time.Since(t0).Seconds())
}()
keyBs := []byte(key)
val, err := s.db.Get(keyBs, nil)
if err == leveldb.ErrNotFound {
databaseOperations.WithLabelValues(dbOpGet, dbResNotFound).Inc()
return DatabaseRecord{}, nil
}
if err != nil {
databaseOperations.WithLabelValues(dbOpGet, dbResError).Inc()
return DatabaseRecord{}, err
}
var rec DatabaseRecord
if err := rec.Unmarshal(val); err != nil {
databaseOperations.WithLabelValues(dbOpGet, dbResUnmarshalError).Inc()
return DatabaseRecord{}, nil
}
rec.Addresses = expire(rec.Addresses, s.clock.Now().UnixNano())
databaseOperations.WithLabelValues(dbOpGet, dbResSuccess).Inc()
return rec, nil
}
func (s *levelDBStore) Serve() {
t := time.NewTimer(0)
defer t.Stop()
defer s.db.Close()
// Start the statistics serve routine. It will exit with us when
// statisticsTrigger is closed.
statisticsTrigger := make(chan struct{})
defer close(statisticsTrigger)
statisticsDone := make(chan struct{})
go s.statisticsServe(statisticsTrigger, statisticsDone)
for {
select {
case fn := <-s.inbox:
// Run function in serialized order.
fn()
case <-t.C:
// Trigger the statistics routine to do its thing in the
// background.
statisticsTrigger <- struct{}{}
case <-statisticsDone:
// The statistics routine is done with one iteratation, schedule
// the next.
t.Reset(databaseStatisticsInterval)
case <-s.stop:
// We're done.
return
}
}
}
func (s *levelDBStore) statisticsServe(trigger <-chan struct{}, done chan<- struct{}) {
for range trigger {
t0 := time.Now()
nowNanos := t0.UnixNano()
cutoff24h := t0.Add(-24 * time.Hour).UnixNano()
cutoff1w := t0.Add(-7 * 24 * time.Hour).UnixNano()
cutoff2Mon := t0.Add(-60 * 24 * time.Hour).UnixNano()
cmd/stdiscosrv: New discovery server (fixes #4618) This is a new revision of the discovery server. Relevant changes and non-changes: - Protocol towards clients is unchanged. - Recommended large scale design is still to be deployed nehind nginx (I tested, and it's still a lot faster at terminating TLS). - Database backend is leveldb again, only. It scales enough, is easy to setup, and we don't need any backend to take care of. - Server supports replication. This is a simple TCP channel - protect it with a firewall when deploying over the internet. (We deploy this within the same datacenter, and with firewall.) Any incoming client announces are sent over the replication channel(s) to other peer discosrvs. Incoming replication changes are applied to the database as if they came from clients, but without the TLS/certificate overhead. - Metrics are exposed using the prometheus library, when enabled. - The database values and replication protocol is protobuf, because JSON was quite CPU intensive when I tried that and benchmarked it. - The "Retry-After" value for failed lookups gets slowly increased from a default of 120 seconds, by 5 seconds for each failed lookup, independently by each discosrv. This lowers the query load over time for clients that are never seen. The Retry-After maxes out at 3600 after a couple of weeks of this increase. The number of failed lookups is stored in the database, now and then (avoiding making each lookup a database put). All in all this means clients can be pointed towards a cluster using just multiple A / AAAA records to gain both load sharing and redundancy (if one is down, clients will talk to the remaining ones). GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4648
2018-01-14 08:52:31 +00:00
current, last24h, last1w, inactive, errors := 0, 0, 0, 0, 0
iter := s.db.NewIterator(&util.Range{}, nil)
for iter.Next() {
// Attempt to unmarshal the record and count the
// failure if there's something wrong with it.
var rec DatabaseRecord
if err := rec.Unmarshal(iter.Value()); err != nil {
errors++
continue
}
// If there are addresses that have not expired it's a current
// record, otherwise account it based on when it was last seen
// (last 24 hours or last week) or finally as inactice.
switch {
case len(expire(rec.Addresses, nowNanos)) > 0:
current++
case rec.Seen > cutoff24h:
last24h++
case rec.Seen > cutoff1w:
last1w++
case rec.Seen > cutoff2Mon:
inactive++
case rec.Missed < cutoff2Mon:
// It hasn't been seen lately and we haven't recorded
// someone asking for this device in a long time either;
// delete the record.
if err := s.db.Delete(iter.Key(), nil); err != nil {
databaseOperations.WithLabelValues(dbOpDelete, dbResError).Inc()
} else {
databaseOperations.WithLabelValues(dbOpDelete, dbResSuccess).Inc()
}
cmd/stdiscosrv: New discovery server (fixes #4618) This is a new revision of the discovery server. Relevant changes and non-changes: - Protocol towards clients is unchanged. - Recommended large scale design is still to be deployed nehind nginx (I tested, and it's still a lot faster at terminating TLS). - Database backend is leveldb again, only. It scales enough, is easy to setup, and we don't need any backend to take care of. - Server supports replication. This is a simple TCP channel - protect it with a firewall when deploying over the internet. (We deploy this within the same datacenter, and with firewall.) Any incoming client announces are sent over the replication channel(s) to other peer discosrvs. Incoming replication changes are applied to the database as if they came from clients, but without the TLS/certificate overhead. - Metrics are exposed using the prometheus library, when enabled. - The database values and replication protocol is protobuf, because JSON was quite CPU intensive when I tried that and benchmarked it. - The "Retry-After" value for failed lookups gets slowly increased from a default of 120 seconds, by 5 seconds for each failed lookup, independently by each discosrv. This lowers the query load over time for clients that are never seen. The Retry-After maxes out at 3600 after a couple of weeks of this increase. The number of failed lookups is stored in the database, now and then (avoiding making each lookup a database put). All in all this means clients can be pointed towards a cluster using just multiple A / AAAA records to gain both load sharing and redundancy (if one is down, clients will talk to the remaining ones). GitHub-Pull-Request: https://github.com/syncthing/syncthing/pull/4648
2018-01-14 08:52:31 +00:00
default:
inactive++
}
}
iter.Release()
databaseKeys.WithLabelValues("current").Set(float64(current))
databaseKeys.WithLabelValues("last24h").Set(float64(last24h))
databaseKeys.WithLabelValues("last1w").Set(float64(last1w))
databaseKeys.WithLabelValues("inactive").Set(float64(inactive))
databaseKeys.WithLabelValues("error").Set(float64(errors))
databaseStatisticsSeconds.Set(time.Since(t0).Seconds())
// Signal that we are done and can be scheduled again.
done <- struct{}{}
}
}
func (s *levelDBStore) Stop() {
close(s.stop)
}
// merge returns the merged result of the two database records a and b. The
// result is the union of the two address sets, with the newer expiry time
// chosen for any duplicates.
func merge(a, b DatabaseRecord) DatabaseRecord {
// Both lists must be sorted for this to work.
sort.Slice(a.Addresses, func(i, j int) bool {
return a.Addresses[i].Address < a.Addresses[j].Address
})
sort.Slice(b.Addresses, func(i, j int) bool {
return b.Addresses[i].Address < b.Addresses[j].Address
})
res := DatabaseRecord{
Addresses: make([]DatabaseAddress, 0, len(a.Addresses)+len(b.Addresses)),
Seen: a.Seen,
}
if b.Seen > a.Seen {
res.Seen = b.Seen
}
aIdx := 0
bIdx := 0
aAddrs := a.Addresses
bAddrs := b.Addresses
loop:
for {
switch {
case aIdx == len(aAddrs) && bIdx == len(bAddrs):
// both lists are exhausted, we are done
break loop
case aIdx == len(aAddrs):
// a is exhausted, pick from b and continue
res.Addresses = append(res.Addresses, bAddrs[bIdx])
bIdx++
continue
case bIdx == len(bAddrs):
// b is exhausted, pick from a and continue
res.Addresses = append(res.Addresses, aAddrs[aIdx])
aIdx++
continue
}
// We have values left on both sides.
aVal := aAddrs[aIdx]
bVal := bAddrs[bIdx]
switch {
case aVal.Address == bVal.Address:
// update for same address, pick newer
if aVal.Expires > bVal.Expires {
res.Addresses = append(res.Addresses, aVal)
} else {
res.Addresses = append(res.Addresses, bVal)
}
aIdx++
bIdx++
case aVal.Address < bVal.Address:
// a is smallest, pick it and continue
res.Addresses = append(res.Addresses, aVal)
aIdx++
default:
// b is smallest, pick it and continue
res.Addresses = append(res.Addresses, bVal)
bIdx++
}
}
return res
}
// expire returns the list of addresses after removing expired entries.
// Expiration happen in place, so the slice given as the parameter is
// destroyed. Internal order is not preserved.
func expire(addrs []DatabaseAddress, now int64) []DatabaseAddress {
i := 0
for i < len(addrs) {
if addrs[i].Expires < now {
// This item is expired. Replace it with the last in the list
// (noop if we are at the last item).
addrs[i] = addrs[len(addrs)-1]
// Wipe the last item of the list to release references to
// strings and stuff.
addrs[len(addrs)-1] = DatabaseAddress{}
// Shorten the slice.
addrs = addrs[:len(addrs)-1]
continue
}
i++
}
return addrs
}