contrib: Add cluster persistence docs

And remove sections from the INTERNALS.md page as they are replaced.
This commit is contained in:
Daniel Nephin 2021-07-05 15:56:48 -04:00
parent faa06040e4
commit 2b8c80f03e
5 changed files with 138 additions and 22 deletions

View File

@ -30,16 +30,6 @@ The components in this section are shared between Consul agents in client and se
| [agent/router](https://github.com/hashicorp/consul/tree/main/agent/router), [agent/pool](https://github.com/hashicorp/consul/tree/main/agent/pool) | These are used for routing RPC queries to Consul servers and for connection pooling. |
| [agent/structs](https://github.com/hashicorp/consul/tree/main/agent/structs) | This has definitions of all the internal RPC protocol request and response structures. |
### Server Components
The components in this section are only used by Consul servers.
| Directory | Contents |
| --------- | -------- |
| [agent/consul](https://github.com/hashicorp/consul/tree/main/agent/consul) | This is where the Consul server object is defined, and the top-level `consul` package has all of the functionality that's used by server agents. This includes things like the internal RPC endpoints. |
| [agent/consul/fsm](https://github.com/hashicorp/consul/tree/main/agent/consul/fsm), [agent/consul/state](https://github.com/hashicorp/consul/tree/main/agent/consul/state) | These components make up Consul's finite state machine (updated by the Raft consensus algorithm) and backed by the state store (based on immutable radix trees). All updates of Consul's consistent state is handled by the finite state machine, and all read queries to the Consul servers are serviced by the state store's data structures. |
| [agent/consul/autopilot](https://github.com/hashicorp/consul/tree/main/agent/consul/autopilot) | This contains a package of functions that provide Consul's [Autopilot](https://www.consul.io/docs/guides/autopilot.html) features. |
### Other Components
There are several other top-level packages used internally by Consul as well as externally by other applications.
@ -48,7 +38,5 @@ There are several other top-level packages used internally by Consul as well as
| --------- | -------- |
| [acl](https://github.com/hashicorp/consul/tree/main/api) | This supports the underlying policy engine for Consul's [ACL](https://www.consul.io/docs/guides/acl.html) system. |
| [api](https://github.com/hashicorp/consul/tree/main/api) | This `api` package provides an official Go API client for Consul, which is also used by Consul's [CLI](https://www.consul.io/docs/commands/index.html) commands to communicate with the local Consul agent. |
| [command](https://github.com/hashicorp/consul/tree/main/command) | This contains a sub-package for each of Consul's [CLI](https://www.consul.io/docs/commands/index.html) command implementations. |
| [snapshot](https://github.com/hashicorp/consul/tree/main/snapshot) | This has implementation details for Consul's [snapshot archives](https://www.consul.io/api/snapshot.html). |
| [api/watch](https://github.com/hashicorp/consul/tree/main/api/watch) | This has implementation details for Consul's [watches](https://www.consul.io/docs/agent/watches.html), used both internally to Consul and by the [watch CLI command](https://www.consul.io/docs/commands/watch.html). |
| [website](https://github.com/hashicorp/consul/tree/main/website) | This has the full source code for [consul.io](https://www.consul.io/). Pull requests can update the source code and Consul's documentation all together. |

View File

@ -1,4 +1,4 @@
{
"theme": "default",
"themeCSS": ".node rect { fill: rgb(220, 71, 125); stroke-width: 1; stroke: black; } .node .label { color: white; } "
"themeCSS": ".node rect { fill: rgb(220, 71, 125); stroke-width: 1; stroke: black; } .node .label { color: white; }; .cluster rect { fill: #f0f0f0; stroke-width: 1px; stroke: #333}; .edgeLabel { background-color: #f0f0f0; }"
}

View File

@ -1,15 +1,108 @@
# Persistence
# Cluster Persistence
- hashicorp/raft
- state.Store - hashicorp/go-memdb
- FSM
- boltdb - https://github.com/boltdb/bolt (https://github.com/etcd-io/bbolt)
- snapshot and restore
The cluser persistence subsystem runs entirely in Server Agents. It handles both read and
write requests from the [RPC] subsystem. See the [Consul Architecture Guide] for an
introduction to the Consul deployment architecture and the [Consensus Protocol] used by
the cluster persistence subsystem.
[RPC]: ../rpc
[Consul Architecture Guide]: https://www.consul.io/docs/architecture
[Consensus Protocol]: https://www.consul.io/docs/architecture/consensus
## Additional documentation
![Overview](./overview.svg)
The following links provide more detail about about cluster persistence.
<sup>[source](./overview.mmd)</sup>
* [Consul Architecture Guide](https://www.consul.io/docs/architecture)
## Raft and FSM
[hashicorp/raft] is at the core of cluster persistence. Raft requires an [FSM], a
finite-state machine implementation, to persist state changes. The Consul FSM is
implemented in [agent/consul/fsm] as a set of commands.
[FSM]: https://pkg.go.dev/github.com/hashicorp/raft#FSM
[hashicorp/raft]: https://github.com/hashicorp/raft
[agent/consul/fsm]: https://github.com/hashicorp/consul/tree/main/agent/consul/fsm
Raft also requires a [LogStore] to persist logs to disk. Consul uses [hashicorp/raft-boltdb]
which implements [LogStore] using [boltdb]. In the near future we should be updating to
use [bbolt].
[LogStore]: https://pkg.go.dev/github.com/hashicorp/raft#LogStore
[hashicorp/raft-boltdb]: https://github.com/hashicorp/raft-boltdb
[boltdb]: https://github.com/boltdb/bolt
[bbolt]: https://github.com/etcd-io/bbolt
## State Store
Consul stores the full state of the cluster in memory using the state store. The state store is
implemented in [agent/consul/state] and uses [hashicorp/go-memdb] to maintain indexes of
data stored in a set of tables. The main entrypoint to the state store is [NewStateStore].
[agent/consul/state]: https://github.com/hashicorp/consul/tree/main/agent/consul/state
[hashicorp/go-memdb]: https://github.com/hashicorp/go-memdb
[NewStateStore]: https://github.com/hashicorp/consul/blob/main/agent/consul/state/state_store.go
### Tables, Schemas, and Indexes
The state store is organized as a set of tables, and each table has a set of indexes.
`newDBSchema` in [schema.go] shows the full list of tables, and each schema function shows
the full list of indexes.
[schema.go]: https://github.com/hashicorp/consul/blob/main/agent/consul/state/schema.go
There are two styles for defining table indexes. The original style uses generic indexer
implementations from [hashicorp/go-memdb] (ex: `StringFieldIndex`). These indexes use
[reflect] to find values for an index. These generic indexers work well when the index
value is a single value available directly from the struct field, and there are no
oss/enterprise differences.
The second style of indexers are custom indexers implemented using only functions and
based on the types defined in [indexer.go]. This style of index works well when the index
value is a value derived from one or multiple fields, or when there are oss/enterprise
differences between the indexes.
[reflect]: https://golang.org/pkg/reflect/
[indexer.go]: https://github.com/hashicorp/consul/blob/main/agent/consul/state/indexer.go
## Snapshot and Restore
Snapshots are the primary mechanism used to backup the data stored by cluster persistence.
If all Consul servers fail, a snapshot can be used to restore the cluster back
to its previous state.
Note that there are two different snapshot and restore concepts that exist at different
layers. First there is the `Snapshot` and `Restore` methods on the raft [FSM] interface,
that Consul must implement. These methods are implemented as mostly passthrough to the
state store. These methods may be called internally by raft to perform log compaction
(snapshot) or to bootstrap a new follower (restore). Consul implements snapshot and
restore using the `Snapshot` and `Restore` types in [agent/consul/state].
Snapshot and restore also exist as actions that a user may perform. There are [CLI]
commands, [HTTP API] endpoints, and [RPC] endpoints that allow a user to capture an
archive which contains a snapshot of the state, and restore that state to a running
cluster. The [consul/snapshot] package provides some of the logic for creating and reading
the snapshot archives for users. See [commands/snapshot] for a reference to these user
facing operations.
[CLI]: ../cli
[HTTP API]: ../http-api
[commands/snapshot]: https://www.consul.io/commands/snapshot
[consul/snapshot]: https://github.com/hashicorp/consul/tree/main/snapshot
Finally, there is also a [snapshot agent] (enterprise only) that uses the snapshot API
endpoints to periodically capture a snapshot, and optionally send it somewhere for
storage.
[snapshot agent]: https://www.consul.io/commands/snapshot/agent
## Raft Autopilot
[hashicorp/raft-autopilot] is used by Consul to automate some parts of the upgrade process.
[hashicorp/raft-autopilot]: https://github.com/hashicorp/raft-autopilot

View File

@ -0,0 +1,34 @@
graph TB
requestLeader[request] --> RPCLeader
requestFollower[request] --> RPCFollower
class requestLeader,requestFollower req;
classDef req fill:transparent,color:#000,stroke-width:1;
subgraph Leader
RPCLeader[RPC]
RaftLeader[Raft]
StateStoreLeader[State Store]
FSMLeader[FSM]
end
RPCLeader -->|write| RaftLeader
RPCLeader -->|read| StateStoreLeader
RaftLeader ---> FSMLeader
FSMLeader --> StateStoreLeader
subgraph Follower
RPCFollower[RPC]
RaftFollower[Raft]
StateStoreFollower[State Store]
FSMFollower[FSM]
end
RaftLeader <-.->|consensus and replication| RaftFollower
RPCFollower -->|forward write to leader| RPCLeader
RPCFollower -->|read| StateStoreFollower
RaftFollower --> FSMFollower
FSMFollower --> StateStoreFollower

File diff suppressed because one or more lines are too long

After

Width:  |  Height:  |  Size: 22 KiB