| 1 | Simple-IPC API |
| 2 | ============== |
| 3 | |
| 4 | The Simple-IPC API is a collection of `ipc_` prefixed library routines |
| 5 | and a basic communication protocol that allows an IPC-client process to |
| 6 | send an application-specific IPC-request message to an IPC-server |
| 7 | process and receive an application-specific IPC-response message. |
| 8 | |
| 9 | Communication occurs over a named pipe on Windows and a Unix domain |
| 10 | socket on other platforms. IPC-clients and IPC-servers rendezvous at |
| 11 | a previously agreed-to application-specific pathname (which is outside |
| 12 | the scope of this design) that is local to the computer system. |
| 13 | |
| 14 | The IPC-server routines within the server application process create a |
| 15 | thread pool to listen for connections and receive request messages |
| 16 | from multiple concurrent IPC-clients. When received, these messages |
| 17 | are dispatched up to the server application callbacks for handling. |
| 18 | IPC-server routines then incrementally relay responses back to the |
| 19 | IPC-client. |
| 20 | |
| 21 | The IPC-client routines within a client application process connect |
| 22 | to the IPC-server and send a request message and wait for a response. |
| 23 | When received, the response is returned back to the caller. |
| 24 | |
| 25 | For example, the `fsmonitor--daemon` feature will be built as a server |
| 26 | application on top of the IPC-server library routines. It will have |
| 27 | threads watching for file system events and a thread pool waiting for |
| 28 | client connections. Clients, such as `git status`, will request a list |
| 29 | of file system events since a point in time and the server will |
| 30 | respond with a list of changed files and directories. The formats of |
| 31 | the request and response are application-specific; the IPC-client and |
| 32 | IPC-server routines treat them as opaque byte streams. |
| 33 | |
| 34 | |
| 35 | Comparison with sub-process model |
| 36 | --------------------------------- |
| 37 | |
| 38 | The Simple-IPC mechanism differs from the existing `sub-process.c` |
| 39 | model (Documentation/technical/long-running-process-protocol.adoc) and |
| 40 | used by applications like Git-LFS. In the LFS-style sub-process model, |
| 41 | the helper is started by the foreground process, communication happens |
| 42 | via a pair of file descriptors bound to the stdin/stdout of the |
| 43 | sub-process, the sub-process only serves the current foreground |
| 44 | process, and the sub-process exits when the foreground process |
| 45 | terminates. |
| 46 | |
| 47 | In the Simple-IPC model the server is a very long-running service. It |
| 48 | can service many clients at the same time and has a private socket or |
| 49 | named pipe connection to each active client. It might be started |
| 50 | (on-demand) by the current client process or it might have been |
| 51 | started by a previous client or by the OS at boot time. The server |
| 52 | process is not associated with a terminal and it persists after |
| 53 | clients terminate. Clients do not have access to the stdin/stdout of |
| 54 | the server process and therefore must communicate over sockets or |
| 55 | named pipes. |
| 56 | |
| 57 | |
| 58 | Server startup and shutdown |
| 59 | --------------------------- |
| 60 | |
| 61 | How an application server based upon IPC-server is started is also |
| 62 | outside the scope of the Simple-IPC design and is a property of the |
| 63 | application using it. For example, the server might be started or |
| 64 | restarted during routine maintenance operations, or it might be |
| 65 | started as a system service during the system boot-up sequence, or it |
| 66 | might be started on-demand by a foreground Git command when needed. |
| 67 | |
| 68 | Similarly, server shutdown is a property of the application using |
| 69 | the simple-ipc routines. For example, the server might decide to |
| 70 | shutdown when idle or only upon explicit request. |
| 71 | |
| 72 | |
| 73 | Simple-IPC protocol |
| 74 | ------------------- |
| 75 | |
| 76 | The Simple-IPC protocol consists of a single request message from the |
| 77 | client and an optional response message from the server. Both the |
| 78 | client and server messages are unlimited in length and are terminated |
| 79 | with a flush packet. |
| 80 | |
| 81 | The pkt-line routines (linkgit:gitprotocol-common[5]) |
| 82 | are used to simplify buffer management during message generation, |
| 83 | transmission, and reception. A flush packet is used to mark the end |
| 84 | of the message. This allows the sender to incrementally generate and |
| 85 | transmit the message. It allows the receiver to incrementally receive |
| 86 | the message in chunks and to know when they have received the entire |
| 87 | message. |
| 88 | |
| 89 | The actual byte format of the client request and server response |
| 90 | messages are application specific. The IPC layer transmits and |
| 91 | receives them as opaque byte buffers without any concern for the |
| 92 | content within. It is the job of the calling application layer to |
| 93 | understand the contents of the request and response messages. |
| 94 | |
| 95 | |
| 96 | Summary |
| 97 | ------- |
| 98 | |
| 99 | Conceptually, the Simple-IPC protocol is similar to an HTTP REST |
| 100 | request. Clients connect, make an application-specific and |
| 101 | stateless request, receive an application-specific |
| 102 | response, and disconnect. It is a one round trip facility for |
| 103 | querying the server. The Simple-IPC routines hide the socket, |
| 104 | named pipe, and thread pool details and allow the application |
| 105 | layer to focus on the task at hand. |