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 01:52:31 -07:00
|
|
|
// 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/.
|
2014-06-01 13:50:14 -07:00
|
|
|
|
2013-12-22 19:35:05 -07:00
|
|
|
package main
|
|
|
|
|
|
|
|
import (
|
2015-09-13 02:44:33 -07:00
|
|
|
"crypto/tls"
|
2014-02-20 09:40:15 -07:00
|
|
|
"flag"
|
2013-12-22 19:35:05 -07:00
|
|
|
"log"
|
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 01:52:31 -07:00
|
|
|
"net"
|
|
|
|
"net/http"
|
2014-02-20 09:40:15 -07:00
|
|
|
"os"
|
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 01:52:31 -07:00
|
|
|
"strings"
|
2014-02-17 01:23:37 -07:00
|
|
|
"time"
|
2013-12-22 19:35:05 -07:00
|
|
|
|
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 01:52:31 -07:00
|
|
|
"github.com/prometheus/client_golang/prometheus/promhttp"
|
2019-10-07 04:30:25 -07:00
|
|
|
"github.com/syncthing/syncthing/lib/build"
|
2015-11-04 09:55:21 -07:00
|
|
|
"github.com/syncthing/syncthing/lib/protocol"
|
2016-08-22 23:41:49 -07:00
|
|
|
"github.com/syncthing/syncthing/lib/tlsutil"
|
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 01:52:31 -07:00
|
|
|
"github.com/syndtr/goleveldb/leveldb/opt"
|
2018-09-08 02:56:56 -07:00
|
|
|
"github.com/thejerf/suture"
|
2013-12-22 19:35:05 -07:00
|
|
|
)
|
|
|
|
|
2015-11-13 02:14:19 -07:00
|
|
|
const (
|
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 01:52:31 -07:00
|
|
|
addressExpiryTime = 2 * time.Hour
|
|
|
|
databaseStatisticsInterval = 5 * time.Minute
|
|
|
|
|
|
|
|
// Reannounce-After is set to reannounceAfterSeconds +
|
|
|
|
// random(reannounzeFuzzSeconds), similar for Retry-After
|
|
|
|
reannounceAfterSeconds = 3300
|
|
|
|
reannounzeFuzzSeconds = 300
|
|
|
|
errorRetryAfterSeconds = 1500
|
|
|
|
errorRetryFuzzSeconds = 300
|
|
|
|
|
|
|
|
// Retry for not found is minSeconds + failures * incSeconds +
|
|
|
|
// random(fuzz), where failures is the number of consecutive lookups
|
|
|
|
// with no answer, up to maxSeconds. The fuzz is applied after capping
|
|
|
|
// to maxSeconds.
|
|
|
|
notFoundRetryMinSeconds = 60
|
|
|
|
notFoundRetryMaxSeconds = 3540
|
|
|
|
notFoundRetryIncSeconds = 10
|
|
|
|
notFoundRetryFuzzSeconds = 60
|
|
|
|
|
|
|
|
// How often (in requests) we serialize the missed counter to database.
|
|
|
|
notFoundMissesWriteInterval = 10
|
|
|
|
|
|
|
|
httpReadTimeout = 5 * time.Second
|
|
|
|
httpWriteTimeout = 5 * time.Second
|
|
|
|
httpMaxHeaderBytes = 1 << 10
|
|
|
|
|
|
|
|
// Size of the replication outbox channel
|
|
|
|
replicationOutboxSize = 10000
|
2015-11-13 02:14:19 -07:00
|
|
|
)
|
|
|
|
|
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 01:52:31 -07:00
|
|
|
// These options make the database a little more optimized for writes, at
|
|
|
|
// the expense of some memory usage and risk of losing writes in a (system)
|
|
|
|
// crash.
|
|
|
|
var levelDBOptions = &opt.Options{
|
|
|
|
NoSync: true,
|
|
|
|
WriteBuffer: 32 << 20, // default 4<<20
|
|
|
|
}
|
|
|
|
|
2013-12-22 19:35:05 -07:00
|
|
|
var (
|
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 01:52:31 -07:00
|
|
|
debug = false
|
2013-12-22 19:35:05 -07:00
|
|
|
)
|
|
|
|
|
|
|
|
func main() {
|
2014-02-20 09:40:15 -07:00
|
|
|
var listen string
|
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 01:52:31 -07:00
|
|
|
var dir string
|
|
|
|
var metricsListen string
|
|
|
|
var replicationListen string
|
|
|
|
var replicationPeers string
|
|
|
|
var certFile string
|
|
|
|
var keyFile string
|
|
|
|
var useHTTP bool
|
2015-03-25 00:16:52 -07:00
|
|
|
|
|
|
|
log.SetOutput(os.Stdout)
|
|
|
|
log.SetFlags(0)
|
2014-02-20 09:40:15 -07:00
|
|
|
|
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 01:52:31 -07:00
|
|
|
flag.StringVar(&certFile, "cert", "./cert.pem", "Certificate file")
|
|
|
|
flag.StringVar(&dir, "db-dir", "./discovery.db", "Database directory")
|
|
|
|
flag.BoolVar(&debug, "debug", false, "Print debug output")
|
|
|
|
flag.BoolVar(&useHTTP, "http", false, "Listen on HTTP (behind an HTTPS proxy)")
|
2015-09-13 02:44:33 -07:00
|
|
|
flag.StringVar(&listen, "listen", ":8443", "Listen address")
|
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 01:52:31 -07:00
|
|
|
flag.StringVar(&keyFile, "key", "./key.pem", "Key file")
|
|
|
|
flag.StringVar(&metricsListen, "metrics-listen", "", "Metrics listen address")
|
|
|
|
flag.StringVar(&replicationPeers, "replicate", "", "Replication peers, id@address, comma separated")
|
|
|
|
flag.StringVar(&replicationListen, "replication-listen", ":19200", "Replication listen address")
|
2019-10-07 04:30:25 -07:00
|
|
|
showVersion := flag.Bool("version", false, "Show version")
|
2014-02-20 09:40:15 -07:00
|
|
|
flag.Parse()
|
|
|
|
|
2020-04-16 01:09:33 -07:00
|
|
|
log.Println(build.LongVersionFor("stdiscosrv"))
|
2019-10-07 04:30:25 -07:00
|
|
|
if *showVersion {
|
|
|
|
return
|
|
|
|
}
|
2016-06-02 04:58:39 -07:00
|
|
|
|
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 01:52:31 -07:00
|
|
|
cert, err := tls.LoadX509KeyPair(certFile, keyFile)
|
2020-03-19 06:50:24 -07:00
|
|
|
if os.IsNotExist(err) {
|
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 01:52:31 -07:00
|
|
|
log.Println("Failed to load keypair. Generating one, this might take a while...")
|
2019-10-16 11:31:46 -07:00
|
|
|
cert, err = tlsutil.NewCertificate(certFile, keyFile, "stdiscosrv", 20*365)
|
2015-11-06 09:36:59 -07:00
|
|
|
if err != nil {
|
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 01:52:31 -07:00
|
|
|
log.Fatalln("Failed to generate X509 key pair:", err)
|
|
|
|
}
|
2020-03-19 06:50:24 -07:00
|
|
|
} else if err != nil {
|
|
|
|
log.Fatalln("Failed to load keypair:", err)
|
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 01:52:31 -07:00
|
|
|
}
|
|
|
|
devID := protocol.NewDeviceID(cert.Certificate[0])
|
|
|
|
log.Println("Server device ID is", devID)
|
|
|
|
|
|
|
|
// Parse the replication specs, if any.
|
|
|
|
var allowedReplicationPeers []protocol.DeviceID
|
|
|
|
var replicationDestinations []string
|
|
|
|
parts := strings.Split(replicationPeers, ",")
|
|
|
|
for _, part := range parts {
|
|
|
|
fields := strings.Split(part, "@")
|
|
|
|
|
|
|
|
switch len(fields) {
|
|
|
|
case 2:
|
|
|
|
// This is an id@address specification. Grab the address for the
|
|
|
|
// destination list. Try to resolve it once to catch obvious
|
|
|
|
// syntax errors here rather than having the sender service fail
|
|
|
|
// repeatedly later.
|
|
|
|
_, err := net.ResolveTCPAddr("tcp", fields[1])
|
2016-08-22 23:41:49 -07:00
|
|
|
if err != nil {
|
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 01:52:31 -07:00
|
|
|
log.Fatalln("Resolving address:", err)
|
2016-08-22 23:41:49 -07:00
|
|
|
}
|
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 01:52:31 -07:00
|
|
|
replicationDestinations = append(replicationDestinations, fields[1])
|
|
|
|
fallthrough // N.B.
|
2013-12-22 19:35:05 -07:00
|
|
|
|
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 01:52:31 -07:00
|
|
|
case 1:
|
|
|
|
// The first part is always a device ID.
|
|
|
|
id, err := protocol.DeviceIDFromString(fields[0])
|
|
|
|
if err != nil {
|
|
|
|
log.Fatalln("Parsing device ID:", err)
|
|
|
|
}
|
|
|
|
allowedReplicationPeers = append(allowedReplicationPeers, id)
|
|
|
|
|
|
|
|
default:
|
|
|
|
log.Fatalln("Unrecognized replication spec:", part)
|
|
|
|
}
|
2015-11-06 09:36:59 -07:00
|
|
|
}
|
2015-11-04 09:55:21 -07:00
|
|
|
|
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 01:52:31 -07:00
|
|
|
// Root of the service tree.
|
2018-08-13 11:39:08 -07:00
|
|
|
main := suture.New("main", suture.Spec{
|
2018-09-08 02:56:56 -07:00
|
|
|
PassThroughPanics: true,
|
2018-08-13 11:39:08 -07:00
|
|
|
})
|
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 01:52:31 -07:00
|
|
|
|
|
|
|
// Start the database.
|
|
|
|
db, err := newLevelDBStore(dir)
|
2014-09-08 02:48:26 -07:00
|
|
|
if err != nil {
|
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 01:52:31 -07:00
|
|
|
log.Fatalln("Open database:", err)
|
2014-09-08 02:48:26 -07:00
|
|
|
}
|
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 01:52:31 -07:00
|
|
|
main.Add(db)
|
|
|
|
|
|
|
|
// Start any replication senders.
|
|
|
|
var repl replicationMultiplexer
|
|
|
|
for _, dst := range replicationDestinations {
|
|
|
|
rs := newReplicationSender(dst, cert, allowedReplicationPeers)
|
|
|
|
main.Add(rs)
|
|
|
|
repl = append(repl, rs)
|
2014-04-19 14:14:56 -07:00
|
|
|
}
|
2014-02-17 01:23:37 -07:00
|
|
|
|
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 01:52:31 -07:00
|
|
|
// If we have replication configured, start the replication listener.
|
|
|
|
if len(allowedReplicationPeers) > 0 {
|
|
|
|
rl := newReplicationListener(replicationListen, cert, allowedReplicationPeers, db)
|
|
|
|
main.Add(rl)
|
|
|
|
}
|
2014-02-20 09:40:15 -07:00
|
|
|
|
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 01:52:31 -07:00
|
|
|
// Start the main API server.
|
|
|
|
qs := newAPISrv(listen, cert, db, repl, useHTTP)
|
|
|
|
main.Add(qs)
|
|
|
|
|
|
|
|
// If we have a metrics port configured, start a metrics handler.
|
|
|
|
if metricsListen != "" {
|
|
|
|
go func() {
|
|
|
|
mux := http.NewServeMux()
|
|
|
|
mux.Handle("/metrics", promhttp.Handler())
|
|
|
|
log.Fatal(http.ListenAndServe(metricsListen, mux))
|
|
|
|
}()
|
2014-06-29 16:42:03 -07:00
|
|
|
}
|
2014-04-03 13:44:40 -07:00
|
|
|
|
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 01:52:31 -07:00
|
|
|
// Engage!
|
|
|
|
main.Serve()
|
2014-09-08 02:48:26 -07:00
|
|
|
}
|