From e54effec45642942c961a96222dbbfa8d901a8f3 Mon Sep 17 00:00:00 2001 From: Alex Auvolat Date: Mon, 19 Feb 2024 11:36:22 +0100 Subject: [PATCH 01/13] [doc-fixes] fixes to index of configuration options --- doc/book/reference-manual/configuration.md | 23 ++++++++++------------ 1 file changed, 10 insertions(+), 13 deletions(-) diff --git a/doc/book/reference-manual/configuration.md b/doc/book/reference-manual/configuration.md index 5e12a7da..02064444 100644 --- a/doc/book/reference-manual/configuration.md +++ b/doc/book/reference-manual/configuration.md @@ -83,7 +83,7 @@ Top-level configuration options: [`block_size`](#block_size), [`bootstrap_peers`](#bootstrap_peers), [`compression_level`](#compression_level), -[`data_dir`](#metadata_dir), +[`data_dir`](#data_dir), [`data_fsync`](#data_fsync), [`db_engine`](#db_engine), [`lmdb_map_size`](#lmdb_map_size), @@ -92,20 +92,19 @@ Top-level configuration options: [`replication_mode`](#replication_mode), [`rpc_bind_addr`](#rpc_bind_addr), [`rpc_public_addr`](#rpc_public_addr), -[`rpc_secret`](#rpc_secret), -[`rpc_secret_file`](#rpc_secret), +[`rpc_secret`/`rpc_secret_file`](#rpc_secret), [`sled_cache_capacity`](#sled_cache_capacity), [`sled_flush_every_ms`](#sled_flush_every_ms). The `[consul_discovery]` section: [`api`](#consul_api), [`ca_cert`](#consul_ca_cert), -[`client_cert`](#consul_client_cert), -[`client_key`](#consul_client_cert), +[`client_cert`](#consul_client_cert_and_key), +[`client_key`](#consul_client_cert_and_key), [`consul_http_addr`](#consul_http_addr), -[`meta`](#consul_tags), +[`meta`](#consul_tags_and_meta), [`service_name`](#consul_service_name), -[`tags`](#consul_tags), +[`tags`](#consul_tags_and_meta), [`tls_skip_verify`](#consul_tls_skip_verify), [`token`](#consul_token). @@ -125,10 +124,8 @@ The `[s3_web]` section: The `[admin]` section: [`api_bind_addr`](#admin_api_bind_addr), -[`metrics_token`](#admin_metrics_token), -[`metrics_token_file`](#admin_metrics_token), -[`admin_token`](#admin_token), -[`admin_token_file`](#admin_token), +[`metrics_token`/`metrics_token_file`](#admin_metrics_token), +[`admin_token`/`admin_token_file`](#admin_token), [`trace_sink`](#admin_trace_sink), @@ -474,7 +471,7 @@ the `/v1/catalog` endpoints, enabling mTLS if `client_cert` and `client_key` are `service_name` should be set to the service name under which Garage's RPC ports are announced. -#### `client_cert`, `client_key` {#consul_client_cert} +#### `client_cert`, `client_key` {#consul_client_cert_and_key} TLS client certificate and client key to use when communicating with Consul over TLS. Both are mandatory when doing so. Only available when `api = "catalog"`. @@ -508,7 +505,7 @@ node_prefix "" { } ``` -#### `tags` and `meta` {#consul_tags} +#### `tags` and `meta` {#consul_tags_and_meta} Additional list of tags and map of service meta to add during service registration. From e73cb79e1edf9324c4880c1c92e4f9c869b8b130 Mon Sep 17 00:00:00 2001 From: Alex Auvolat Date: Mon, 19 Feb 2024 11:42:06 +0100 Subject: [PATCH 02/13] [doc-fixes] configuration reference: fix typo and set block size in M --- doc/book/reference-manual/configuration.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/doc/book/reference-manual/configuration.md b/doc/book/reference-manual/configuration.md index 02064444..27d9fd54 100644 --- a/doc/book/reference-manual/configuration.md +++ b/doc/book/reference-manual/configuration.md @@ -17,7 +17,7 @@ data_fsync = false db_engine = "lmdb" -block_size = 1048576 +block_size = "1M" sled_cache_capacity = "128MiB" sled_flush_every_ms = 2000 @@ -31,7 +31,7 @@ rpc_public_addr = "[fc00:1::1]:3901" bootstrap_peers = [ "563e1ac825ee3323aa441e72c26d1030d6d4414aeb3dd25287c531e7fc2bc95d@[fc00:1::1]:3901", - "86f0f26ae4afbd59aaf9cfb059eefac844951efd5b8caeec0d53f4ed6c85f332[fc00:1::2]:3901", + "86f0f26ae4afbd59aaf9cfb059eefac844951efd5b8caeec0d53f4ed6c85f332@[fc00:1::2]:3901", "681456ab91350f92242e80a531a3ec9392cb7c974f72640112f90a600d7921a4@[fc00:B::1]:3901", "212fd62eeaca72c122b45a7f4fa0f55e012aa5e24ac384a72a3016413fa724ff@[fc00:F::1]:3901", ] From f0bbad2db95b00ec429f498fe15f1007bd87da5e Mon Sep 17 00:00:00 2001 From: Alex Auvolat Date: Mon, 19 Feb 2024 10:58:54 +0100 Subject: [PATCH 03/13] [networking-fixes] use rpc_public_addr in netapp's HelloMessage --- src/net/netapp.rs | 19 ++++++++++++------- src/rpc/system.rs | 10 +++++----- 2 files changed, 17 insertions(+), 12 deletions(-) diff --git a/src/net/netapp.rs b/src/net/netapp.rs index b1ad9db8..e572dc11 100644 --- a/src/net/netapp.rs +++ b/src/net/netapp.rs @@ -38,6 +38,11 @@ pub(crate) type VersionTag = [u8; 16]; /// Value of the Netapp version used in the version tag pub(crate) const NETAPP_VERSION_TAG: u64 = 0x6e65746170700005; // netapp 0x0005 +/// HelloMessage is sent by the client on a Netapp connection to indicate +/// that they are also a server and ready to recieve incoming connections +/// at the specified address and port. If the client doesn't know their +/// public address, they don't need to specify it and we look at the +/// remote address of the socket is used instead. #[derive(Serialize, Deserialize, Debug)] pub(crate) struct HelloMessage { pub server_addr: Option, @@ -56,9 +61,6 @@ type OnDisconnectHandler = Box; /// NetApp can be used in a stand-alone fashion or together with a peering strategy. /// If using it alone, you will want to set `on_connect` and `on_disconnect` events /// in order to manage information about the current peer list. -/// -/// It is generally not necessary to use NetApp stand-alone, as the provided full mesh -/// and RPS peering strategies take care of the most common use cases. pub struct NetApp { listen_params: ArcSwapOption, @@ -83,7 +85,7 @@ pub struct NetApp { struct ListenParams { listen_addr: SocketAddr, - public_addr: Option, + public_addr: Option, } impl NetApp { @@ -180,7 +182,7 @@ impl NetApp { pub async fn listen( self: Arc, listen_addr: SocketAddr, - public_addr: Option, + public_addr: Option, mut must_exit: watch::Receiver, ) { let listen_params = ListenParams { @@ -396,8 +398,11 @@ impl NetApp { } if let Some(lp) = self.listen_params.load_full() { - let server_addr = lp.public_addr; - let server_port = lp.listen_addr.port(); + let server_addr = lp.public_addr.map(|x| x.ip()); + let server_port = lp + .public_addr + .map(|x| x.port()) + .unwrap_or(lp.listen_addr.port()); let hello_endpoint = self.hello_endpoint.load_full().unwrap(); tokio::spawn(async move { hello_endpoint diff --git a/src/rpc/system.rs b/src/rpc/system.rs index de44e656..5c6e07b1 100644 --- a/src/rpc/system.rs +++ b/src/rpc/system.rs @@ -98,7 +98,6 @@ pub struct System { system_endpoint: Arc>, rpc_listen_addr: SocketAddr, - #[cfg(any(feature = "consul-discovery", feature = "kubernetes-discovery"))] rpc_public_addr: Option, bootstrap_peers: Vec, @@ -369,7 +368,6 @@ impl System { replication_mode, replication_factor, rpc_listen_addr: config.rpc_bind_addr, - #[cfg(any(feature = "consul-discovery", feature = "kubernetes-discovery"))] rpc_public_addr, bootstrap_peers: config.bootstrap_peers.clone(), #[cfg(feature = "consul-discovery")] @@ -390,9 +388,11 @@ impl System { /// Perform bootstraping, starting the ping loop pub async fn run(self: Arc, must_exit: watch::Receiver) { join!( - self.netapp - .clone() - .listen(self.rpc_listen_addr, None, must_exit.clone()), + self.netapp.clone().listen( + self.rpc_listen_addr, + self.rpc_public_addr, + must_exit.clone() + ), self.peering.clone().run(must_exit.clone()), self.discovery_loop(must_exit.clone()), self.status_exchange_loop(must_exit.clone()), From b96f84b894684ed43e281a3aa2f391b424414a84 Mon Sep 17 00:00:00 2001 From: Alex Auvolat Date: Mon, 19 Feb 2024 11:24:33 +0100 Subject: [PATCH 04/13] [networking-fixes] add option to bind outgoing RPC sockets (fix #638) Thanks to yuka for the original patch. --- src/garage/main.rs | 2 +- src/net/netapp.rs | 26 ++++++++++++++++++++++---- src/net/test.rs | 2 +- src/rpc/system.rs | 5 ++++- src/util/config.rs | 3 +++ 5 files changed, 31 insertions(+), 7 deletions(-) diff --git a/src/garage/main.rs b/src/garage/main.rs index 4d8dcc67..e489fff0 100644 --- a/src/garage/main.rs +++ b/src/garage/main.rs @@ -203,7 +203,7 @@ async fn cli_command(opt: Opt) -> Result<(), Error> { // Generate a temporary keypair for our RPC client let (_pk, sk) = sodiumoxide::crypto::sign::ed25519::gen_keypair(); - let netapp = NetApp::new(GARAGE_VERSION_TAG, network_key, sk); + let netapp = NetApp::new(GARAGE_VERSION_TAG, network_key, sk, None); // Find and parse the address of the target host let (id, addr, is_default_addr) = if let Some(h) = opt.rpc_host { diff --git a/src/net/netapp.rs b/src/net/netapp.rs index e572dc11..faa51a99 100644 --- a/src/net/netapp.rs +++ b/src/net/netapp.rs @@ -13,7 +13,7 @@ use sodiumoxide::crypto::sign::ed25519; use futures::stream::futures_unordered::FuturesUnordered; use futures::stream::StreamExt; -use tokio::net::{TcpListener, TcpStream}; +use tokio::net::{TcpListener, TcpSocket, TcpStream}; use tokio::select; use tokio::sync::{mpsc, watch}; @@ -62,6 +62,7 @@ type OnDisconnectHandler = Box; /// If using it alone, you will want to set `on_connect` and `on_disconnect` events /// in order to manage information about the current peer list. pub struct NetApp { + bind_outgoing_to: Option, listen_params: ArcSwapOption, /// Version tag, 8 bytes for netapp version, 8 bytes for app version @@ -94,13 +95,19 @@ impl NetApp { /// using `.listen()` /// /// Our Peer ID is the public key associated to the secret key given here. - pub fn new(app_version_tag: u64, netid: auth::Key, privkey: ed25519::SecretKey) -> Arc { + pub fn new( + app_version_tag: u64, + netid: auth::Key, + privkey: ed25519::SecretKey, + bind_outgoing_to: Option, + ) -> Arc { let mut version_tag = [0u8; 16]; version_tag[0..8].copy_from_slice(&u64::to_be_bytes(NETAPP_VERSION_TAG)[..]); version_tag[8..16].copy_from_slice(&u64::to_be_bytes(app_version_tag)[..]); let id = privkey.public_key(); let netapp = Arc::new(Self { + bind_outgoing_to, listen_params: ArcSwapOption::new(None), version_tag, netid, @@ -300,9 +307,20 @@ impl NetApp { return Ok(()); } - let socket = TcpStream::connect(ip).await?; + let stream = match self.bind_outgoing_to { + Some(addr) => { + let socket = if addr.is_ipv4() { + TcpSocket::new_v4()? + } else { + TcpSocket::new_v6()? + }; + socket.bind(SocketAddr::new(addr, 0))?; + socket.connect(ip).await? + } + None => TcpStream::connect(ip).await?, + }; info!("Connected to {}, negotiating handshake...", ip); - ClientConn::init(self, socket, id).await?; + ClientConn::init(self, stream, id).await?; Ok(()) } diff --git a/src/net/test.rs b/src/net/test.rs index c6259752..5a3f236d 100644 --- a/src/net/test.rs +++ b/src/net/test.rs @@ -102,7 +102,7 @@ fn run_netapp( Arc, Arc, ) { - let netapp = NetApp::new(0u64, netid, sk); + let netapp = NetApp::new(0u64, netid, sk, None); let peering = PeeringManager::new(netapp.clone(), bootstrap_peers, None); let peering2 = peering.clone(); diff --git a/src/rpc/system.rs b/src/rpc/system.rs index 5c6e07b1..147ec4d6 100644 --- a/src/rpc/system.rs +++ b/src/rpc/system.rs @@ -324,7 +324,10 @@ impl System { warn!("This Garage node does not know its publicly reachable RPC address, this might hamper intra-cluster communication."); } - let netapp = NetApp::new(GARAGE_VERSION_TAG, network_key, node_key); + let bind_outgoing_to = Some(config) + .filter(|x| x.rpc_bind_outgoing) + .map(|x| x.rpc_bind_addr.ip()); + let netapp = NetApp::new(GARAGE_VERSION_TAG, network_key, node_key, bind_outgoing_to); let peering = PeeringManager::new(netapp.clone(), vec![], rpc_public_addr); if let Some(ping_timeout) = config.rpc_ping_timeout_msec { peering.set_ping_timeout_millis(ping_timeout); diff --git a/src/util/config.rs b/src/util/config.rs index a9a72110..056c625d 100644 --- a/src/util/config.rs +++ b/src/util/config.rs @@ -55,6 +55,9 @@ pub struct Config { pub rpc_secret_file: Option, /// Address to bind for RPC pub rpc_bind_addr: SocketAddr, + /// Bind outgoing sockets to rpc_bind_addr's IP address as well + #[serde(default)] + pub rpc_bind_outgoing: bool, /// Public IP address of this node pub rpc_public_addr: Option, From eab54b37981073651712a15334b60b2c409f2576 Mon Sep 17 00:00:00 2001 From: Alex Auvolat Date: Mon, 19 Feb 2024 11:31:09 +0100 Subject: [PATCH 05/13] [networking-fixes] add doc for rpc_bind_outgoing --- doc/book/reference-manual/configuration.md | 13 +++++++++++++ 1 file changed, 13 insertions(+) diff --git a/doc/book/reference-manual/configuration.md b/doc/book/reference-manual/configuration.md index 27d9fd54..f1474613 100644 --- a/doc/book/reference-manual/configuration.md +++ b/doc/book/reference-manual/configuration.md @@ -27,6 +27,7 @@ compression_level = 1 rpc_secret = "4425f5c26c5e11581d3223904324dcb5b5d5dfb14e5e7f35e38c595424f5f1e6" rpc_bind_addr = "[::]:3901" +rpc_bind_outgoing = false rpc_public_addr = "[fc00:1::1]:3901" bootstrap_peers = [ @@ -91,6 +92,7 @@ Top-level configuration options: [`metadata_fsync`](#metadata_fsync), [`replication_mode`](#replication_mode), [`rpc_bind_addr`](#rpc_bind_addr), +[`rpc_bind_outgoing`](#rpc_bind_outgoing), [`rpc_public_addr`](#rpc_public_addr), [`rpc_secret`/`rpc_secret_file`](#rpc_secret), [`sled_cache_capacity`](#sled_cache_capacity), @@ -415,6 +417,17 @@ the node, even in the case of a NAT: the NAT should be configured to forward the port number to the same internal port nubmer. This means that if you have several nodes running behind a NAT, they should each use a different RPC port number. +#### `rpc_bind_outgoing` {#rpc_bind_outgoing} (since v0.9.2) + +If enabled, pre-bind all sockets for outgoing connections to the same IP address +used for listening (the IP address specified in `rpc_bind_addr`) before +trying to connect to remote nodes. +This can be necessary if a node has multiple IP addresses, +but only one is allowed or able to reach the other nodes, +for instance due to firewall rules or specific routing configuration. + +Disabled by default. + #### `rpc_public_addr` {#rpc_public_addr} The address and port that other nodes need to use to contact this node for From 66fe893023a1c296deeb7a4d4c82336c4e34103f Mon Sep 17 00:00:00 2001 From: Alex Auvolat Date: Mon, 19 Feb 2024 17:52:53 +0100 Subject: [PATCH 06/13] [networking-fixes] garage_net: retry connecting when new IP is learned --- src/net/peering.rs | 28 ++++++++++++++++++++-------- 1 file changed, 20 insertions(+), 8 deletions(-) diff --git a/src/net/peering.rs b/src/net/peering.rs index 32199cf8..3f0b2279 100644 --- a/src/net/peering.rs +++ b/src/net/peering.rs @@ -80,6 +80,23 @@ impl PeerInfoInternal { failed_pings: 0, } } + fn add_addr(&mut self, addr: SocketAddr) -> bool { + if !self.all_addrs.contains(&addr) { + self.all_addrs.push(addr); + // If we are learning a new address for this node, + // we want to retry connecting + self.state = match self.state { + PeerConnState::Trying(_) => PeerConnState::Trying(0), + PeerConnState::Waiting(_, _) | PeerConnState::Abandonned => { + PeerConnState::Waiting(0, Instant::now()) + } + x @ (PeerConnState::Ourself | PeerConnState::Connected) => x, + }; + true + } else { + false + } + } } /// Information that the full mesh peering strategy can return about the peers it knows of @@ -465,8 +482,7 @@ impl PeeringManager { let mut changed = false; for (id, addr) in list.iter() { if let Some(kh) = known_hosts.list.get_mut(id) { - if !kh.all_addrs.contains(addr) { - kh.all_addrs.push(*addr); + if kh.add_addr(*addr) { changed = true; } } else { @@ -538,9 +554,7 @@ impl PeeringManager { let mut known_hosts = self.known_hosts.write().unwrap(); if is_incoming { if let Some(host) = known_hosts.list.get_mut(&id) { - if !host.all_addrs.contains(&addr) { - host.all_addrs.push(addr); - } + host.add_addr(addr); } else { known_hosts.list.insert(id, self.new_peer(&id, addr)); } @@ -553,9 +567,7 @@ impl PeeringManager { if let Some(host) = known_hosts.list.get_mut(&id) { host.state = PeerConnState::Connected; host.addr = addr; - if !host.all_addrs.contains(&addr) { - host.all_addrs.push(addr); - } + host.add_addr(addr); } else { known_hosts .list From 95e3a39b4d0fe6de551fac61e9f88ce488e41477 Mon Sep 17 00:00:00 2001 From: Alex Auvolat Date: Mon, 19 Feb 2024 18:07:35 +0100 Subject: [PATCH 07/13] [networking-fixes] small refactoring in garage_net peering --- src/net/peering.rs | 26 ++++++++++++-------------- 1 file changed, 12 insertions(+), 14 deletions(-) diff --git a/src/net/peering.rs b/src/net/peering.rs index 3f0b2279..61882a18 100644 --- a/src/net/peering.rs +++ b/src/net/peering.rs @@ -164,23 +164,22 @@ struct KnownHosts { impl KnownHosts { fn new() -> Self { let list = HashMap::new(); - let hash = Self::calculate_hash(&list); + let hash = Self::calculate_hash(vec![]); Self { list, hash } } fn update_hash(&mut self) { - self.hash = Self::calculate_hash(&self.list); + self.hash = Self::calculate_hash(self.connected_peers_vec()); } - fn map_into_vec(input: &HashMap) -> Vec<(NodeID, SocketAddr)> { - let mut list = Vec::with_capacity(input.len()); - for (id, peer) in input.iter() { - if peer.state == PeerConnState::Connected || peer.state == PeerConnState::Ourself { + fn connected_peers_vec(&self) -> Vec<(NodeID, SocketAddr)> { + let mut list = Vec::with_capacity(self.list.len()); + for (id, peer) in self.list.iter() { + if peer.state.is_up() { list.push((*id, peer.addr)); } } list } - fn calculate_hash(input: &HashMap) -> hash::Digest { - let mut list = Self::map_into_vec(input); + fn calculate_hash(mut list: Vec<(NodeID, SocketAddr)>) -> hash::Digest { list.sort(); let mut hash_state = hash::State::new(); for (id, addr) in list { @@ -231,6 +230,7 @@ impl PeeringManager { netapp.id, PeerInfoInternal::new(addr, PeerConnState::Ourself), ); + known_hosts.update_hash(); } // TODO for v0.10 / v1.0 : rename the endpoint (it will break compatibility) @@ -251,13 +251,11 @@ impl PeeringManager { let strat2 = strat.clone(); netapp.on_connected(move |id: NodeID, addr: SocketAddr, is_incoming: bool| { - let strat2 = strat2.clone(); strat2.on_connected(id, addr, is_incoming); }); let strat2 = strat.clone(); netapp.on_disconnected(move |id: NodeID, is_incoming: bool| { - let strat2 = strat2.clone(); strat2.on_disconnected(id, is_incoming); }); @@ -462,7 +460,7 @@ impl PeeringManager { } async fn exchange_peers(self: Arc, id: &NodeID) { - let peer_list = KnownHosts::map_into_vec(&self.known_hosts.read().unwrap().list); + let peer_list = self.known_hosts.read().unwrap().connected_peers_vec(); let pex_message = PeerListMessage { list: peer_list }; match self .peer_list_endpoint @@ -550,7 +548,7 @@ impl PeeringManager { } } - fn on_connected(self: Arc, id: NodeID, addr: SocketAddr, is_incoming: bool) { + fn on_connected(self: &Arc, id: NodeID, addr: SocketAddr, is_incoming: bool) { let mut known_hosts = self.known_hosts.write().unwrap(); if is_incoming { if let Some(host) = known_hosts.list.get_mut(&id) { @@ -578,7 +576,7 @@ impl PeeringManager { self.update_public_peer_list(&known_hosts); } - fn on_disconnected(self: Arc, id: NodeID, is_incoming: bool) { + fn on_disconnected(self: &Arc, id: NodeID, is_incoming: bool) { if !is_incoming { info!("Connection to {} was closed", hex::encode(&id[..8])); let mut known_hosts = self.known_hosts.write().unwrap(); @@ -620,7 +618,7 @@ impl EndpointHandler for PeeringManager { _from: NodeID, ) -> PeerListMessage { self.handle_peer_list(&peer_list.list[..]); - let peer_list = KnownHosts::map_into_vec(&self.known_hosts.read().unwrap().list); + let peer_list = self.known_hosts.read().unwrap().connected_peers_vec(); PeerListMessage { list: peer_list } } } From e91576677e712c07cf9c47b1a0d2cc4d2d1d37cf Mon Sep 17 00:00:00 2001 From: Alex Auvolat Date: Fri, 16 Feb 2024 10:50:41 +0100 Subject: [PATCH 08/13] [reconnect-only-current] filter nodes to reconnect to do not try reconnecting to nodes received from consul/kubernetes discovery if they are not currently in the layout --- src/rpc/system.rs | 21 ++++++++++++++++----- 1 file changed, 16 insertions(+), 5 deletions(-) diff --git a/src/rpc/system.rs b/src/rpc/system.rs index de44e656..14a101ca 100644 --- a/src/rpc/system.rs +++ b/src/rpc/system.rs @@ -725,15 +725,18 @@ impl System { async fn discovery_loop(self: &Arc, mut stop_signal: watch::Receiver) { while !*stop_signal.borrow() { - let not_configured = self.ring.borrow().layout.check().is_err(); - let no_peers = self.peering.get_peer_list().len() < self.replication_factor; - let expected_n_nodes = self.ring.borrow().layout.num_nodes(); - let bad_peers = self + let n_connected = self .peering .get_peer_list() .iter() .filter(|p| p.is_up()) - .count() != expected_n_nodes; + .count(); + + let not_configured = self.ring.borrow().layout.check().is_err(); + let no_peers = n_connected < self.replication_factor; + + let expected_n_nodes = self.ring.borrow().layout.num_nodes(); + let bad_peers = n_connected != expected_n_nodes; if not_configured || no_peers || bad_peers { info!("Doing a bootstrap/discovery step (not_configured: {}, no_peers: {}, bad_peers: {})", not_configured, no_peers, bad_peers); @@ -780,6 +783,14 @@ impl System { } } + if !not_configured && !no_peers { + // If the layout is configured, and we already have some connections + // to other nodes in the cluster, we can skip trying to connect to + // nodes that are not in the cluster layout. + let ring = self.ring.borrow(); + ping_list.retain(|(id, _)| ring.layout.node_ids().contains(&(*id).into())); + } + for (node_id, node_addr) in ping_list { let self2 = self.clone(); tokio::spawn(async move { From 00d479358d31b445bfbe6d7ee3c37520be7e6d85 Mon Sep 17 00:00:00 2001 From: Alex Auvolat Date: Tue, 20 Feb 2024 11:35:18 +0100 Subject: [PATCH 09/13] [peer-metrics] refactor/simplify SystemMetrics --- src/rpc/system.rs | 77 +++++++++++---------------------------- src/rpc/system_metrics.rs | 52 ++++++++++---------------- 2 files changed, 41 insertions(+), 88 deletions(-) diff --git a/src/rpc/system.rs b/src/rpc/system.rs index 147ec4d6..4a505f58 100644 --- a/src/rpc/system.rs +++ b/src/rpc/system.rs @@ -3,11 +3,9 @@ use std::collections::HashMap; use std::io::{Read, Write}; use std::net::{IpAddr, SocketAddr}; use std::path::{Path, PathBuf}; -use std::sync::atomic::Ordering; use std::sync::{Arc, RwLock}; use std::time::{Duration, Instant}; -use arc_swap::ArcSwap; use async_trait::async_trait; use futures::join; use serde::{Deserialize, Serialize}; @@ -88,7 +86,7 @@ pub struct System { persist_cluster_layout: Persister, persist_peer_list: Persister, - local_status: ArcSwap, + local_status: Arc>, node_status: RwLock>, pub netapp: Arc, @@ -106,7 +104,7 @@ pub struct System { #[cfg(feature = "kubernetes-discovery")] kubernetes_discovery: Option, - metrics: SystemMetrics, + _metrics: SystemMetrics, replication_mode: ReplicationMode, replication_factor: usize, @@ -280,10 +278,11 @@ impl System { } }; - let metrics = SystemMetrics::new(replication_factor); - let mut local_status = NodeStatus::initial(replication_factor, &cluster_layout); - local_status.update_disk_usage(&config.metadata_dir, &config.data_dir, &metrics); + local_status.update_disk_usage(&config.metadata_dir, &config.data_dir); + let local_status = Arc::new(RwLock::new(local_status)); + + let metrics = SystemMetrics::new(replication_factor, local_status.clone()); let ring = Ring::new(cluster_layout, replication_factor); let (update_ring, ring) = watch::channel(Arc::new(ring)); @@ -357,7 +356,7 @@ impl System { id: netapp.id.into(), persist_cluster_layout, persist_peer_list, - local_status: ArcSwap::new(Arc::new(local_status)), + local_status, node_status: RwLock::new(HashMap::new()), netapp: netapp.clone(), peering: peering.clone(), @@ -377,7 +376,7 @@ impl System { consul_discovery, #[cfg(feature = "kubernetes-discovery")] kubernetes_discovery: config.kubernetes_discovery.clone(), - metrics, + _metrics: metrics, ring, update_ring: Mutex::new(update_ring), @@ -546,12 +545,9 @@ impl System { } }; + let hostname = self.local_status.read().unwrap().hostname.clone(); if let Err(e) = c - .publish_consul_service( - self.netapp.id, - &self.local_status.load_full().hostname, - rpc_public_addr, - ) + .publish_consul_service(self.netapp.id, &hostname, rpc_public_addr) .await { error!("Error while publishing Consul service: {}", e); @@ -573,13 +569,8 @@ impl System { } }; - if let Err(e) = publish_kubernetes_node( - k, - self.netapp.id, - &self.local_status.load_full().hostname, - rpc_public_addr, - ) - .await + let hostname = self.local_status.read().unwrap().hostname.clone(); + if let Err(e) = publish_kubernetes_node(k, self.netapp.id, &hostname, rpc_public_addr).await { error!("Error while publishing node to Kubernetes: {}", e); } @@ -596,15 +587,13 @@ impl System { } fn update_local_status(&self) { - let mut new_si: NodeStatus = self.local_status.load().as_ref().clone(); + let mut local_status = self.local_status.write().unwrap(); let ring = self.ring.borrow(); - new_si.cluster_layout_version = ring.layout.version; - new_si.cluster_layout_staging_hash = ring.layout.staging_hash; + local_status.cluster_layout_version = ring.layout.version; + local_status.cluster_layout_staging_hash = ring.layout.staging_hash; - new_si.update_disk_usage(&self.metadata_dir, &self.data_dir, &self.metrics); - - self.local_status.swap(Arc::new(new_si)); + local_status.update_disk_usage(&self.metadata_dir, &self.data_dir); } // --- RPC HANDLERS --- @@ -629,7 +618,7 @@ impl System { from: Uuid, info: &NodeStatus, ) -> Result { - let local_info = self.local_status.load(); + let local_info = self.local_status.read().unwrap(); if local_info.replication_factor < info.replication_factor { error!("Some node have a higher replication factor ({}) than this one ({}). This is not supported and will lead to data corruption. Shutting down for safety.", @@ -644,6 +633,8 @@ impl System { tokio::spawn(self.clone().pull_cluster_layout(from)); } + drop(local_info); + self.node_status .write() .unwrap() @@ -708,7 +699,7 @@ impl System { let restart_at = Instant::now() + STATUS_EXCHANGE_INTERVAL; self.update_local_status(); - let local_status: NodeStatus = self.local_status.load().as_ref().clone(); + let local_status: NodeStatus = self.local_status.read().unwrap().clone(); let _ = self .rpc .broadcast( @@ -893,12 +884,7 @@ impl NodeStatus { } } - fn update_disk_usage( - &mut self, - meta_dir: &Path, - data_dir: &DataDirEnum, - metrics: &SystemMetrics, - ) { + fn update_disk_usage(&mut self, meta_dir: &Path, data_dir: &DataDirEnum) { use nix::sys::statvfs::statvfs; let mount_avail = |path: &Path| match statvfs(path) { Ok(x) => { @@ -934,27 +920,6 @@ impl NodeStatus { ) })(), }; - - if let Some((avail, total)) = self.meta_disk_avail { - metrics - .values - .meta_disk_avail - .store(avail, Ordering::Relaxed); - metrics - .values - .meta_disk_total - .store(total, Ordering::Relaxed); - } - if let Some((avail, total)) = self.data_disk_avail { - metrics - .values - .data_disk_avail - .store(avail, Ordering::Relaxed); - metrics - .values - .data_disk_total - .store(total, Ordering::Relaxed); - } } } diff --git a/src/rpc/system_metrics.rs b/src/rpc/system_metrics.rs index af81b71f..7b390c25 100644 --- a/src/rpc/system_metrics.rs +++ b/src/rpc/system_metrics.rs @@ -1,31 +1,22 @@ -use std::sync::atomic::{AtomicU64, Ordering}; -use std::sync::Arc; +use std::sync::{Arc, RwLock}; use opentelemetry::{global, metrics::*, KeyValue}; +use crate::system::NodeStatus; + /// TableMetrics reference all counter used for metrics pub struct SystemMetrics { pub(crate) _garage_build_info: ValueObserver, pub(crate) _replication_factor: ValueObserver, pub(crate) _disk_avail: ValueObserver, pub(crate) _disk_total: ValueObserver, - pub(crate) values: Arc, -} - -#[derive(Default)] -pub struct SystemMetricsValues { - pub(crate) data_disk_total: AtomicU64, - pub(crate) data_disk_avail: AtomicU64, - pub(crate) meta_disk_total: AtomicU64, - pub(crate) meta_disk_avail: AtomicU64, } impl SystemMetrics { - pub fn new(replication_factor: usize) -> Self { + pub fn new(replication_factor: usize, local_status: Arc>) -> Self { let meter = global::meter("garage_system"); - let values = Arc::new(SystemMetricsValues::default()); - let values1 = values.clone(); - let values2 = values.clone(); + let st1 = local_status.clone(); + let st2 = local_status.clone(); Self { _garage_build_info: meter .u64_value_observer("garage_build_info", move |observer| { @@ -47,31 +38,28 @@ impl SystemMetrics { .init(), _disk_avail: meter .u64_value_observer("garage_local_disk_avail", move |observer| { - match values1.data_disk_avail.load(Ordering::Relaxed) { - 0 => (), - x => observer.observe(x, &[KeyValue::new("volume", "data")]), - }; - match values1.meta_disk_avail.load(Ordering::Relaxed) { - 0 => (), - x => observer.observe(x, &[KeyValue::new("volume", "metadata")]), - }; + let st = st1.read().unwrap(); + if let Some((avail, _total)) = st.data_disk_avail { + observer.observe(avail, &[KeyValue::new("volume", "data")]); + } + if let Some((avail, _total)) = st.meta_disk_avail { + observer.observe(avail, &[KeyValue::new("volume", "metadata")]); + } }) .with_description("Garage available disk space on each node") .init(), _disk_total: meter .u64_value_observer("garage_local_disk_total", move |observer| { - match values2.data_disk_total.load(Ordering::Relaxed) { - 0 => (), - x => observer.observe(x, &[KeyValue::new("volume", "data")]), - }; - match values2.meta_disk_total.load(Ordering::Relaxed) { - 0 => (), - x => observer.observe(x, &[KeyValue::new("volume", "metadata")]), - }; + let st = st2.read().unwrap(); + if let Some((_avail, total)) = st.data_disk_avail { + observer.observe(total, &[KeyValue::new("volume", "data")]); + } + if let Some((_avail, total)) = st.meta_disk_avail { + observer.observe(total, &[KeyValue::new("volume", "metadata")]); + } }) .with_description("Garage total disk space on each node") .init(), - values, } } } From 3cdf69f07924d120c572577495789774535daafe Mon Sep 17 00:00:00 2001 From: Alex Auvolat Date: Tue, 20 Feb 2024 12:37:55 +0100 Subject: [PATCH 10/13] [peer-metrics] Add metrics for cluster health, like GetClusterHealth admin API --- src/rpc/system.rs | 12 ++- src/rpc/system_metrics.rs | 188 +++++++++++++++++++++++++++++++++----- 2 files changed, 172 insertions(+), 28 deletions(-) diff --git a/src/rpc/system.rs b/src/rpc/system.rs index 4a505f58..8ecefd84 100644 --- a/src/rpc/system.rs +++ b/src/rpc/system.rs @@ -104,7 +104,7 @@ pub struct System { #[cfg(feature = "kubernetes-discovery")] kubernetes_discovery: Option, - _metrics: SystemMetrics, + metrics: SystemMetrics, replication_mode: ReplicationMode, replication_factor: usize, @@ -168,7 +168,7 @@ pub struct ClusterHealth { pub partitions_all_ok: usize, } -#[derive(Debug, Clone, Copy)] +#[derive(Debug, Clone, Copy, Eq, PartialEq)] pub enum ClusterHealthStatus { /// All nodes are available Healthy, @@ -376,7 +376,7 @@ impl System { consul_discovery, #[cfg(feature = "kubernetes-discovery")] kubernetes_discovery: config.kubernetes_discovery.clone(), - _metrics: metrics, + metrics, ring, update_ring: Mutex::new(update_ring), @@ -698,7 +698,13 @@ impl System { while !*stop_signal.borrow() { let restart_at = Instant::now() + STATUS_EXCHANGE_INTERVAL; + // Update local node status that is exchanged. + // Status variables are exported into Prometheus in SystemMetrics, + // so we take the opportunity to also update here the health status + // that is reported in those metrics. self.update_local_status(); + *self.metrics.health.write().unwrap() = Some(self.health()); + let local_status: NodeStatus = self.local_status.read().unwrap().clone(); let _ = self .rpc diff --git a/src/rpc/system_metrics.rs b/src/rpc/system_metrics.rs index 7b390c25..ee07672e 100644 --- a/src/rpc/system_metrics.rs +++ b/src/rpc/system_metrics.rs @@ -2,22 +2,40 @@ use std::sync::{Arc, RwLock}; use opentelemetry::{global, metrics::*, KeyValue}; -use crate::system::NodeStatus; +use crate::system::{ClusterHealth, ClusterHealthStatus, NodeStatus}; /// TableMetrics reference all counter used for metrics pub struct SystemMetrics { + pub(crate) health: Arc>>, + + // Static values pub(crate) _garage_build_info: ValueObserver, pub(crate) _replication_factor: ValueObserver, + + // Disk space values from System::local_status pub(crate) _disk_avail: ValueObserver, pub(crate) _disk_total: ValueObserver, + + // Health report from System::health() + pub(crate) _cluster_healthy: ValueObserver, + pub(crate) _cluster_available: ValueObserver, + pub(crate) _known_nodes: ValueObserver, + pub(crate) _connected_nodes: ValueObserver, + pub(crate) _storage_nodes: ValueObserver, + pub(crate) _storage_nodes_ok: ValueObserver, + pub(crate) _partitions: ValueObserver, + pub(crate) _partitions_quorum: ValueObserver, + pub(crate) _partitions_all_ok: ValueObserver, } impl SystemMetrics { pub fn new(replication_factor: usize, local_status: Arc>) -> Self { let meter = global::meter("garage_system"); - let st1 = local_status.clone(); - let st2 = local_status.clone(); + let health = Arc::new(RwLock::new(None)); Self { + health: health.clone(), + + // Static values _garage_build_info: meter .u64_value_observer("garage_build_info", move |observer| { observer.observe( @@ -36,30 +54,150 @@ impl SystemMetrics { }) .with_description("Garage replication factor setting") .init(), - _disk_avail: meter - .u64_value_observer("garage_local_disk_avail", move |observer| { - let st = st1.read().unwrap(); - if let Some((avail, _total)) = st.data_disk_avail { - observer.observe(avail, &[KeyValue::new("volume", "data")]); - } - if let Some((avail, _total)) = st.meta_disk_avail { - observer.observe(avail, &[KeyValue::new("volume", "metadata")]); + + // Disk space values from System::local_status + _disk_avail: { + let status = local_status.clone(); + meter + .u64_value_observer("garage_local_disk_avail", move |observer| { + let st = status.read().unwrap(); + if let Some((avail, _total)) = st.data_disk_avail { + observer.observe(avail, &[KeyValue::new("volume", "data")]); + } + if let Some((avail, _total)) = st.meta_disk_avail { + observer.observe(avail, &[KeyValue::new("volume", "metadata")]); + } + }) + .with_description("Garage available disk space on each node") + .init() + }, + _disk_total: { + let status = local_status.clone(); + meter + .u64_value_observer("garage_local_disk_total", move |observer| { + let st = status.read().unwrap(); + if let Some((_avail, total)) = st.data_disk_avail { + observer.observe(total, &[KeyValue::new("volume", "data")]); + } + if let Some((_avail, total)) = st.meta_disk_avail { + observer.observe(total, &[KeyValue::new("volume", "metadata")]); + } + }) + .with_description("Garage total disk space on each node") + .init() + }, + + // Health report from System::health() + _cluster_healthy: { + let health = health.clone(); + meter + .u64_value_observer("cluster_healthy", move |observer| { + if let Some(h) = health.read().unwrap().as_ref() { + if h.status == ClusterHealthStatus::Healthy { + observer.observe(1, &[]); + } else { + observer.observe(0, &[]); + } + } + }) + .with_description("Whether all storage nodes are connected") + .init() + }, + _cluster_available: { + let health = health.clone(); + meter.u64_value_observer("cluster_available", move |observer| { + if let Some(h) = health.read().unwrap().as_ref() { + if h.status != ClusterHealthStatus::Unavailable { + observer.observe(1, &[]); + } else { + observer.observe(0, &[]); + } } }) - .with_description("Garage available disk space on each node") - .init(), - _disk_total: meter - .u64_value_observer("garage_local_disk_total", move |observer| { - let st = st2.read().unwrap(); - if let Some((_avail, total)) = st.data_disk_avail { - observer.observe(total, &[KeyValue::new("volume", "data")]); - } - if let Some((_avail, total)) = st.meta_disk_avail { - observer.observe(total, &[KeyValue::new("volume", "metadata")]); - } - }) - .with_description("Garage total disk space on each node") - .init(), + .with_description("Whether all requests can be served, even if some storage nodes are disconnected") + .init() + }, + _known_nodes: { + let health = health.clone(); + meter + .u64_value_observer("cluster_known_nodes", move |observer| { + if let Some(h) = health.read().unwrap().as_ref() { + observer.observe(h.known_nodes as u64, &[]); + } + }) + .with_description("Number of nodes already seen once in the cluster") + .init() + }, + _connected_nodes: { + let health = health.clone(); + meter + .u64_value_observer("cluster_connected_nodes", move |observer| { + if let Some(h) = health.read().unwrap().as_ref() { + observer.observe(h.connected_nodes as u64, &[]); + } + }) + .with_description("Number of nodes currently connected") + .init() + }, + _storage_nodes: { + let health = health.clone(); + meter + .u64_value_observer("cluster_storage_nodes", move |observer| { + if let Some(h) = health.read().unwrap().as_ref() { + observer.observe(h.storage_nodes as u64, &[]); + } + }) + .with_description("Number of storage nodes declared in the current layout") + .init() + }, + _storage_nodes_ok: { + let health = health.clone(); + meter + .u64_value_observer("cluster_storage_nodes_ok", move |observer| { + if let Some(h) = health.read().unwrap().as_ref() { + observer.observe(h.storage_nodes_ok as u64, &[]); + } + }) + .with_description("Number of storage nodes currently connected") + .init() + }, + _partitions: { + let health = health.clone(); + meter + .u64_value_observer("cluster_partitions", move |observer| { + if let Some(h) = health.read().unwrap().as_ref() { + observer.observe(h.partitions as u64, &[]); + } + }) + .with_description("Number of partitions in the layout") + .init() + }, + _partitions_quorum: { + let health = health.clone(); + meter + .u64_value_observer("cluster_partitions_quorum", move |observer| { + if let Some(h) = health.read().unwrap().as_ref() { + observer.observe(h.partitions_quorum as u64, &[]); + } + }) + .with_description( + "Number of partitions for which we have a quorum of connected nodes", + ) + .init() + }, + _partitions_all_ok: { + let health = health.clone(); + meter + .u64_value_observer("cluster_partitions_all_ok", move |observer| { + if let Some(h) = health.read().unwrap().as_ref() { + observer.observe(h.partitions_all_ok as u64, &[]); + } + }) + .with_description( + "Number of partitions for which all storage nodes are connected", + ) + .init() + }, } } } From 182a23cc1207c97922a24182c15e2b228015e8f4 Mon Sep 17 00:00:00 2001 From: Alex Auvolat Date: Tue, 20 Feb 2024 14:20:58 +0100 Subject: [PATCH 11/13] [peer-metrics] refactor SystemMetrics to hold a reference to System --- src/garage/server.rs | 1 + src/rpc/system.rs | 28 +++++---- src/rpc/system_metrics.rs | 126 ++++++++++++++++++++------------------ 3 files changed, 83 insertions(+), 72 deletions(-) diff --git a/src/garage/server.rs b/src/garage/server.rs index 51b06b8e..6323f957 100644 --- a/src/garage/server.rs +++ b/src/garage/server.rs @@ -162,6 +162,7 @@ pub async fn run_server(config_file: PathBuf, secrets: Secrets) -> Result<(), Er info!("Netapp exited"); // Drop all references so that stuff can terminate properly + garage.system.cleanup(); drop(garage); // Await for all background tasks to end diff --git a/src/rpc/system.rs b/src/rpc/system.rs index 8ecefd84..f4fdcace 100644 --- a/src/rpc/system.rs +++ b/src/rpc/system.rs @@ -6,6 +6,7 @@ use std::path::{Path, PathBuf}; use std::sync::{Arc, RwLock}; use std::time::{Duration, Instant}; +use arc_swap::ArcSwapOption; use async_trait::async_trait; use futures::join; use serde::{Deserialize, Serialize}; @@ -86,7 +87,7 @@ pub struct System { persist_cluster_layout: Persister, persist_peer_list: Persister, - local_status: Arc>, + pub(crate) local_status: RwLock, node_status: RwLock>, pub netapp: Arc, @@ -104,10 +105,10 @@ pub struct System { #[cfg(feature = "kubernetes-discovery")] kubernetes_discovery: Option, - metrics: SystemMetrics, + metrics: ArcSwapOption, replication_mode: ReplicationMode, - replication_factor: usize, + pub(crate) replication_factor: usize, /// The ring pub ring: watch::Receiver>, @@ -280,9 +281,6 @@ impl System { let mut local_status = NodeStatus::initial(replication_factor, &cluster_layout); local_status.update_disk_usage(&config.metadata_dir, &config.data_dir); - let local_status = Arc::new(RwLock::new(local_status)); - - let metrics = SystemMetrics::new(replication_factor, local_status.clone()); let ring = Ring::new(cluster_layout, replication_factor); let (update_ring, ring) = watch::channel(Arc::new(ring)); @@ -356,7 +354,7 @@ impl System { id: netapp.id.into(), persist_cluster_layout, persist_peer_list, - local_status, + local_status: RwLock::new(local_status), node_status: RwLock::new(HashMap::new()), netapp: netapp.clone(), peering: peering.clone(), @@ -376,14 +374,19 @@ impl System { consul_discovery, #[cfg(feature = "kubernetes-discovery")] kubernetes_discovery: config.kubernetes_discovery.clone(), - metrics, + metrics: ArcSwapOption::new(None), ring, update_ring: Mutex::new(update_ring), metadata_dir: config.metadata_dir.clone(), data_dir: config.data_dir.clone(), }); + sys.system_endpoint.set_handler(sys.clone()); + + let metrics = SystemMetrics::new(sys.clone()); + sys.metrics.store(Some(Arc::new(metrics))); + Ok(sys) } @@ -401,6 +404,11 @@ impl System { ); } + pub fn cleanup(&self) { + // Break reference cycle + self.metrics.store(None); + } + // ---- Administrative operations (directly available and // also available through RPC) ---- @@ -699,11 +707,7 @@ impl System { let restart_at = Instant::now() + STATUS_EXCHANGE_INTERVAL; // Update local node status that is exchanged. - // Status variables are exported into Prometheus in SystemMetrics, - // so we take the opportunity to also update here the health status - // that is reported in those metrics. self.update_local_status(); - *self.metrics.health.write().unwrap() = Some(self.health()); let local_status: NodeStatus = self.local_status.read().unwrap().clone(); let _ = self diff --git a/src/rpc/system_metrics.rs b/src/rpc/system_metrics.rs index ee07672e..ad4aca2f 100644 --- a/src/rpc/system_metrics.rs +++ b/src/rpc/system_metrics.rs @@ -1,13 +1,12 @@ use std::sync::{Arc, RwLock}; +use std::time::{Duration, Instant}; use opentelemetry::{global, metrics::*, KeyValue}; -use crate::system::{ClusterHealth, ClusterHealthStatus, NodeStatus}; +use crate::system::{ClusterHealthStatus, System}; /// TableMetrics reference all counter used for metrics pub struct SystemMetrics { - pub(crate) health: Arc>>, - // Static values pub(crate) _garage_build_info: ValueObserver, pub(crate) _replication_factor: ValueObserver, @@ -29,12 +28,25 @@ pub struct SystemMetrics { } impl SystemMetrics { - pub fn new(replication_factor: usize, local_status: Arc>) -> Self { + pub fn new(system: Arc) -> Self { let meter = global::meter("garage_system"); - let health = Arc::new(RwLock::new(None)); - Self { - health: health.clone(), + let health_cache = RwLock::new((Instant::now(), system.health())); + let system2 = system.clone(); + let get_health = Arc::new(move || { + { + let cache = health_cache.read().unwrap(); + if cache.0 > Instant::now() - Duration::from_secs(1) { + return cache.1; + } + } + + let health = system2.health(); + *health_cache.write().unwrap() = (Instant::now(), health); + health + }); + + Self { // Static values _garage_build_info: meter .u64_value_observer("garage_build_info", move |observer| { @@ -48,19 +60,22 @@ impl SystemMetrics { }) .with_description("Garage build info") .init(), - _replication_factor: meter - .u64_value_observer("garage_replication_factor", move |observer| { - observer.observe(replication_factor as u64, &[]) - }) - .with_description("Garage replication factor setting") - .init(), + _replication_factor: { + let replication_factor = system.replication_factor; + meter + .u64_value_observer("garage_replication_factor", move |observer| { + observer.observe(replication_factor as u64, &[]) + }) + .with_description("Garage replication factor setting") + .init() + }, // Disk space values from System::local_status _disk_avail: { - let status = local_status.clone(); + let system = system.clone(); meter .u64_value_observer("garage_local_disk_avail", move |observer| { - let st = status.read().unwrap(); + let st = system.local_status.read().unwrap(); if let Some((avail, _total)) = st.data_disk_avail { observer.observe(avail, &[KeyValue::new("volume", "data")]); } @@ -72,10 +87,10 @@ impl SystemMetrics { .init() }, _disk_total: { - let status = local_status.clone(); + let system = system.clone(); meter .u64_value_observer("garage_local_disk_total", move |observer| { - let st = status.read().unwrap(); + let st = system.local_status.read().unwrap(); if let Some((_avail, total)) = st.data_disk_avail { observer.observe(total, &[KeyValue::new("volume", "data")]); } @@ -87,98 +102,90 @@ impl SystemMetrics { .init() }, - // Health report from System::health() + // Health report from System::() _cluster_healthy: { - let health = health.clone(); + let get_health = get_health.clone(); meter .u64_value_observer("cluster_healthy", move |observer| { - if let Some(h) = health.read().unwrap().as_ref() { - if h.status == ClusterHealthStatus::Healthy { - observer.observe(1, &[]); - } else { - observer.observe(0, &[]); - } + let h = get_health(); + if h.status == ClusterHealthStatus::Healthy { + observer.observe(1, &[]); + } else { + observer.observe(0, &[]); } }) .with_description("Whether all storage nodes are connected") .init() }, _cluster_available: { - let health = health.clone(); + let get_health = get_health.clone(); meter.u64_value_observer("cluster_available", move |observer| { - if let Some(h) = health.read().unwrap().as_ref() { - if h.status != ClusterHealthStatus::Unavailable { - observer.observe(1, &[]); - } else { - observer.observe(0, &[]); - } + let h = get_health(); + if h.status != ClusterHealthStatus::Unavailable { + observer.observe(1, &[]); + } else { + observer.observe(0, &[]); } }) .with_description("Whether all requests can be served, even if some storage nodes are disconnected") .init() }, _known_nodes: { - let health = health.clone(); + let get_health = get_health.clone(); meter .u64_value_observer("cluster_known_nodes", move |observer| { - if let Some(h) = health.read().unwrap().as_ref() { - observer.observe(h.known_nodes as u64, &[]); - } + let h = get_health(); + observer.observe(h.known_nodes as u64, &[]); }) .with_description("Number of nodes already seen once in the cluster") .init() }, _connected_nodes: { - let health = health.clone(); + let get_health = get_health.clone(); meter .u64_value_observer("cluster_connected_nodes", move |observer| { - if let Some(h) = health.read().unwrap().as_ref() { - observer.observe(h.connected_nodes as u64, &[]); - } + let h = get_health(); + observer.observe(h.connected_nodes as u64, &[]); }) .with_description("Number of nodes currently connected") .init() }, _storage_nodes: { - let health = health.clone(); + let get_health = get_health.clone(); meter .u64_value_observer("cluster_storage_nodes", move |observer| { - if let Some(h) = health.read().unwrap().as_ref() { - observer.observe(h.storage_nodes as u64, &[]); - } + let h = get_health(); + observer.observe(h.storage_nodes as u64, &[]); }) .with_description("Number of storage nodes declared in the current layout") .init() }, _storage_nodes_ok: { - let health = health.clone(); + let get_health = get_health.clone(); meter .u64_value_observer("cluster_storage_nodes_ok", move |observer| { - if let Some(h) = health.read().unwrap().as_ref() { - observer.observe(h.storage_nodes_ok as u64, &[]); - } + let h = get_health(); + observer.observe(h.storage_nodes_ok as u64, &[]); }) .with_description("Number of storage nodes currently connected") .init() }, _partitions: { - let health = health.clone(); + let get_health = get_health.clone(); meter .u64_value_observer("cluster_partitions", move |observer| { - if let Some(h) = health.read().unwrap().as_ref() { - observer.observe(h.partitions as u64, &[]); - } + let h = get_health(); + observer.observe(h.partitions as u64, &[]); }) .with_description("Number of partitions in the layout") .init() }, _partitions_quorum: { - let health = health.clone(); + let get_health = get_health.clone(); meter .u64_value_observer("cluster_partitions_quorum", move |observer| { - if let Some(h) = health.read().unwrap().as_ref() { - observer.observe(h.partitions_quorum as u64, &[]); - } + let h = get_health(); + observer.observe(h.partitions_quorum as u64, &[]); }) .with_description( "Number of partitions for which we have a quorum of connected nodes", @@ -186,12 +193,11 @@ impl SystemMetrics { .init() }, _partitions_all_ok: { - let health = health.clone(); + let get_health = get_health.clone(); meter .u64_value_observer("cluster_partitions_all_ok", move |observer| { - if let Some(h) = health.read().unwrap().as_ref() { - observer.observe(h.partitions_all_ok as u64, &[]); - } + let h = get_health(); + observer.observe(h.partitions_all_ok as u64, &[]); }) .with_description( "Number of partitions for which all storage nodes are connected", From b868493da9b7cf6e5703ee5d068ae34938fbff34 Mon Sep 17 00:00:00 2001 From: Alex Auvolat Date: Tue, 20 Feb 2024 14:49:08 +0100 Subject: [PATCH 12/13] [peer-metrics] add basic cluster node status metrics (fix #545) --- src/rpc/system_metrics.rs | 94 +++++++++++++++++++++++++++++++++++++++ 1 file changed, 94 insertions(+) diff --git a/src/rpc/system_metrics.rs b/src/rpc/system_metrics.rs index ad4aca2f..ffbef6df 100644 --- a/src/rpc/system_metrics.rs +++ b/src/rpc/system_metrics.rs @@ -3,6 +3,7 @@ use std::time::{Duration, Instant}; use opentelemetry::{global, metrics::*, KeyValue}; +use crate::ring::Ring; use crate::system::{ClusterHealthStatus, System}; /// TableMetrics reference all counter used for metrics @@ -25,6 +26,10 @@ pub struct SystemMetrics { pub(crate) _partitions: ValueObserver, pub(crate) _partitions_quorum: ValueObserver, pub(crate) _partitions_all_ok: ValueObserver, + + // Status report for individual cluster nodes + pub(crate) _layout_node_connected: ValueObserver, + pub(crate) _layout_node_disconnected_time: ValueObserver, } impl SystemMetrics { @@ -204,6 +209,95 @@ impl SystemMetrics { ) .init() }, + + // Status report for individual cluster nodes + _layout_node_connected: { + let system = system.clone(); + meter + .u64_value_observer("cluster_layout_node_connected", move |observer| { + let ring: Arc = system.ring.borrow().clone(); + let nodes = system.get_known_nodes(); + for (id, _, config) in ring.layout.roles.items().iter() { + if let Some(role) = &config.0 { + let mut kv = vec![ + KeyValue::new("id", format!("{:?}", id)), + KeyValue::new("role_zone", role.zone.clone()), + ]; + match role.capacity { + Some(cap) => { + kv.push(KeyValue::new("role_capacity", cap as i64)); + kv.push(KeyValue::new("role_gateway", 0)); + } + None => { + kv.push(KeyValue::new("role_gateway", 1)); + } + } + + let value; + if let Some(node) = nodes.iter().find(|n| n.id == *id) { + value = if node.is_up { 1 } else { 0 }; + // TODO: if we add address and hostname, and those change, we + // get duplicate metrics, due to bad otel aggregation :( + // Can probably be fixed when we upgrade opentelemetry + // kv.push(KeyValue::new("address", node.addr.to_string())); + // kv.push(KeyValue::new( + // "hostname", + // node.status.hostname.clone(), + // )); + } else { + value = 0; + } + + observer.observe(value, &kv); + } + } + }) + .with_description("Connection status for nodes in the cluster layout") + .init() + }, + _layout_node_disconnected_time: { + let system = system.clone(); + meter + .u64_value_observer("cluster_layout_node_disconnected_time", move |observer| { + let ring: Arc = system.ring.borrow().clone(); + let nodes = system.get_known_nodes(); + for (id, _, config) in ring.layout.roles.items().iter() { + if let Some(role) = &config.0 { + let mut kv = vec![ + KeyValue::new("id", format!("{:?}", id)), + KeyValue::new("role_zone", role.zone.clone()), + ]; + match role.capacity { + Some(cap) => { + kv.push(KeyValue::new("role_capacity", cap as i64)); + kv.push(KeyValue::new("role_gateway", 0)); + } + None => { + kv.push(KeyValue::new("role_gateway", 1)); + } + } + + if let Some(node) = nodes.iter().find(|n| n.id == *id) { + // TODO: see comment above + // kv.push(KeyValue::new("address", node.addr.to_string())); + // kv.push(KeyValue::new( + // "hostname", + // node.status.hostname.clone(), + // )); + if node.is_up { + observer.observe(0, &kv); + } else if let Some(secs) = node.last_seen_secs_ago { + observer.observe(secs, &kv); + } + } + } + } + }) + .with_description( + "Time (in seconds) since last connection to nodes in the cluster layout", + ) + .init() + }, } } } From bcd571ef57662a1dc86ef5bb0c055d3f8bd9e3bc Mon Sep 17 00:00:00 2001 From: Alex Auvolat Date: Tue, 20 Feb 2024 14:59:04 +0100 Subject: [PATCH 13/13] [peer-metrics] add documentation for new cluster status metrics --- doc/book/reference-manual/monitoring.md | 106 ++++++++++++++++++++++++ 1 file changed, 106 insertions(+) diff --git a/doc/book/reference-manual/monitoring.md b/doc/book/reference-manual/monitoring.md index 97c533d3..f392c133 100644 --- a/doc/book/reference-manual/monitoring.md +++ b/doc/book/reference-manual/monitoring.md @@ -27,6 +27,112 @@ Exposes the Garage replication factor configured on the node garage_replication_factor 3 ``` +#### `garage_local_disk_avail` and `garage_local_disk_total` (gauge) + +Reports the available and total disk space on each node, for data and metadata separately. + +``` +garage_local_disk_avail{volume="data"} 540341960704 +garage_local_disk_avail{volume="metadata"} 540341960704 +garage_local_disk_total{volume="data"} 763063566336 +garage_local_disk_total{volume="metadata"} 763063566336 +``` + +### Cluster health status metrics + +#### `cluster_healthy` (gauge) + +Whether all storage nodes are connected (0 or 1) + +``` +cluster_healthy 0 +``` + +#### `cluster_available` (gauge) + +Whether all requests can be served, even if some storage nodes are disconnected + +``` +cluster_available 1 +``` + +#### `cluster_connected_nodes` (gauge) + +Number of nodes currently connected + +``` +cluster_connected_nodes 3 +``` + +#### `cluster_known_nodes` (gauge) + +Number of nodes already seen once in the cluster + +``` +cluster_known_nodes 3 +``` + +#### `cluster_layout_node_connected` (gauge) + +Connection status for individual nodes of the cluster layout + +``` +cluster_layout_node_connected{id="62b218d848e86a64",role_capacity="1000000000",role_gateway="0",role_zone="dc1"} 1 +cluster_layout_node_connected{id="a11c7cf18af29737",role_capacity="1000000000",role_gateway="0",role_zone="dc1"} 0 +cluster_layout_node_connected{id="a235ac7695e0c54d",role_capacity="1000000000",role_gateway="0",role_zone="dc1"} 1 +cluster_layout_node_connected{id="b10c110e4e854e5a",role_capacity="1000000000",role_gateway="0",role_zone="dc1"} 1 +``` + +#### `cluster_layout_node_disconnected_time` (gauge) + +Time (in seconds) since last connection to individual nodes of the cluster layout + +``` +cluster_layout_node_disconnected_time{id="62b218d848e86a64",role_capacity="1000000000",role_gateway="0",role_zone="dc1"} 0 +cluster_layout_node_disconnected_time{id="a235ac7695e0c54d",role_capacity="1000000000",role_gateway="0",role_zone="dc1"} 0 +cluster_layout_node_disconnected_time{id="b10c110e4e854e5a",role_capacity="1000000000",role_gateway="0",role_zone="dc1"} 0 +``` + +#### `cluster_storage_nodes` (gauge) + +Number of storage nodes declared in the current layout + +``` +cluster_storage_nodes 4 +``` + +#### `cluster_storage_nodes_ok` (gauge) + +Number of storage nodes currently connected + +``` +cluster_storage_nodes_ok 3 +``` + +#### `cluster_partitions` (gauge) + +Number of partitions in the layout (this is always 256) + +``` +cluster_partitions 256 +``` + +#### `cluster_partitions_all_ok` (gauge) + +Number of partitions for which all storage nodes are connected + +``` +cluster_partitions_all_ok 64 +``` + +#### `cluster_partitions_quorum` (gauge) + +Number of partitions for which we have a quorum of connected nodes and all requests can be served + +``` +cluster_partitions_quorum 256 +``` + ### Metrics of the API endpoints #### `api_admin_request_counter` (counter)