「ピアツーピア層」を編集中
この編集を取り消せます。
下記の差分を確認して、本当に取り消していいか検証してください。よろしければ変更を保存して取り消しを完了してください。
| 最新版 | 編集中の文章 | ||
| 1行目: | 1行目: | ||
| − | .. | + | .. _p2p: |
| − | = | + | = The peer-to-peer layer = |
| − | + | This document explains the inner workings of the peer-to-peer layer of the Tezos shell. This part is in charge of establishing and maintaining network connections with other nodes (gossip). | |
| − | + | The P2P layer is instanciated by the node. It is parametrized by the type of messages that are exchanged over the network (to allow different P2P protocol versions/extensions), and the type of metadata associated to each peer. The latter is useful to compute a score for each peer that reflects the level of trust we have in it. Different policies can be used when communicating with peers with different score values. | |
| − | + | The P2P layer is comprised of a pool of connections, a set of operations on those connections, and a set of workers following the worker pattern pervasively used in the code base. | |
| − | + | The P2P layer is packaged in :package:<code>tezos-p2p</code>, which has documentation for all modules. | |
| − | == | + | == General operation == |
| − | + | I/O Scheduling <sub><sub><sub><sub><sub><sub>~</sub></sub></sub></sub></sub></sub>~ | |
| − | + | The P2P layer uses I/O scheduling in order to be able to control its bandwidth usage as well as implementing different policies (e.g. read/write quotas) to different peers. For now, each peer is granted a fair share of the global allocated bandwidth, but it is planned for the individual allocated bandwidth to each peer to be a function of the peer's score. | |
| − | + | Encryption <sub><sub>~</sub>~</sub>~~ | |
| − | + | The connection between each peer is encrypted using <code>NaCl</code> authenticated-encryption <code>API <http://nacl.cr.yp.to/box.html></code>__. This is done to provide an additional level of security and tamper-proof guarantees in the communication between peers. | |
| − | + | Message queues <sub><sub><sub><sub><sub><sub>~</sub></sub></sub></sub></sub></sub>~ | |
| − | + | On top of basic I/O scheduling, two finite-size typed message queues are used to store incoming (resp. outgoing) messages for each peer. This further restricts the speed at which communication is possible with a peer; when a queue is full, it is not possible to read (resp. write) an additional message. The high-level <code>P2p_socket.connection <../api/odoc/tezos-p2p/Tezos_p2p/P2p_socket/index.html#type-connection></code>__ type by the P2P layer is basically a UNIX socket upgraded with I/O scheduling, peer metadata, cryptographic keys and two messages queues operated by dedicated workers which operate on those queues. | |
| − | + | Pool of connections <sub><sub><sub>~</sub></sub><sub><sub><sub>~</sub></sub></sub>~</sub> | |
| − | + | All the above modules are used in <code>P2p_pool <../api/odoc/tezos-p2p/Tezos_p2p/P2p_pool/index.html></code>__, which constitutes the core of the P2P layer, together with the worker processes described below. It comprises various tables of connections as well as methods to query them, also connections are extended with another message queue where lower level messages (like responses to ping) are filtered out and only application-level messages are kept. | |
| − | + | The main entry point of the P2P layer is in module <code>P2p <../api/odoc/tezos-p2p/Tezos_p2p/P2p/index.html></code>__. See below for a description of workers acting onto the P2P layer. | |
| − | == | + | == Welcome worker == |
| − | + | The welcome worker is responsible for accepting incoming connections and register them into the pool of connections managed by the P2P layer. It basically runs the <code>accept(2)</code> syscall and call <code>P2p_pool.accept <../api/odoc/tezos-p2p/Tezos_p2p/P2p_pool/index.html#val-accept></code>__ so that it is made aware of an incoming connection. From there, the pool will decide how this new connection must be handled. | |
| − | == | + | == Maintenance worker == |
| − | + | The maintenance worker is in charge of establishing an appropriate number of connections with other nodes in order to guarantee a realistic view of the state of the blockchain. It is created with a set of targets to reach regarding the desired amount of peers it needs to keep an active connection to. | |
| − | + | At the pool level, the minimum (resp. maximum) acceptable number of connections is defined. | |
| − | + | At the maintenance worker level, two other sets of thresholds are defined: <code>target</code> (min and max) and <code>threshold</code> (min and max). | |
| − | + | Given these bounds, the maintenance worker: | |
| − | * | + | * Will be triggered every two minutes, when asked by the shell, or when the minimum or maximum number of acceptable connections is reached, whichever happens first. |
| − | * | + | * Will perform the following actions when triggered: if the number of connections is above <code>max_threshold</code>, it will kill connections randomly until it reaches <code>max_target</code> connections. If the number of connections is below <code>min_threshold</code>, it will attempt to connect to peers until it reaches at least <code>min_target</code> connections (and never more than <code>max_target</code> connections). |
