Gluster source code Read 3--MGMT Xlator

Introduction

Personally, Gluster has 5 basic components, namely: Mgmt,rpc SERVER,RPC client, NFS SERVER,NFS client

These 5 components make up the basic services of Gluster, so this chapter is a brief overview of what Mgmt did

Mgmt,nfs,rpc-server Use the same main function (Glusterfsd:main), so all 3 classes have the same loading process, and the approximate flow is as follows:

int main (int argc, char *argv[])

{

.......

CTX = Glusterfs_ctx_new ();

ret = Glusterfs_globals_init ();

ret = Glusterfs_ctx_defaults_init (CTX);

ret = Parse_cmdline (argc, argv, CTX);

ret = Logging_init (CTX);

Gf_proc_dump_init ();

ret = Create_fuse_mount (CTX);

ret = Daemonize (CTX); Become a daemon process

ret = Glusterfs_volumes_init (CTX); It's very important to do something in this function.

ret = Event_dispatch (Ctx->event_pool); Distribution events, reference 3.3

}

From the frequency of CTX, the head can think of CTX is a very important structure, it holds a lot of information, as to what, and then the subsequent chapters to expand, but in this main, all the previous calls are for 2 important functions to pave

1, Glusterfs_volumes_init ()

2, Event_dispatch ()

In Glusterfs_volumes_init () a different xlator (by Volfile-id, which determines which xlator to configure) is configured in Gluster, Xlator is a modular component. The form of so exists, and the storage-side path is

So loading different xlator is equivalent to loading a different GLUSTERFSD service. three different modes

The Gluster has 3 modes of operation, namely:

Gf_server_process,gf_glusterd_process,gf_client_process

Although the 3 commands point to the same file (see Figure 1-1), the file does have 3 modes

Static uint8_t

Gf_get_process_mode (char *exec_name)

{

Char *dup_execname = NULL, *base = NULL;

uint8_t ret = 0;

Dup_execname = Gf_strdup (exec_name);

Base = basename (Dup_execname);

if (!strncmp (base, "GLUSTERFSD", 10)) {

ret = gf_server_process; (Volume,nfs,rpc Server)

} else if (!strncmp (base, "Glusterd", 8)) {

ret = gf_glusterd_process; (Daemon)

} else {

ret = gf_client_process; (RPC Client)

}

Gf_free (Dup_execname);

return ret;

Parse command-line arguments

Using the ARGP package, parse the command line sequence, detailed reference

Http://www.gnu.org/software/libc/manual/html_node/Argp-Examples.html#Argp-Examples

#0 parse_opts (key=16777219, arg=0x0, state=0x7fffffffce70) at glusterfsd.c:780

#1 0x00007ffff62cfc9bin argp_parse () from/lib64/libc.so.6--It's a function of so.

#2 0x0000000000409001in parse_cmdline (argc=6, argv=0x7fffffffe578, ctx=0x617010) atglusterfsd.c:1793

#3 0x000000000040a1dfin Main (argc=6, argv=0x7fffffffe578) at glusterfsd.c:2296

Event_dispatch

was initialized in Glusterfs_ctx_defaults_init.

A lot of function pointers are used in gluster, so how to find the corresponding function of pointers is very helpful to the reading of code, the following 2 structs, the corresponding event function is determined

//event.h

struct Event_ops {

struct Event_pool * (*new) (int count,int eventthreadcount);

Int (*event_register) (structevent_pool *event_pool, int fd,

Event_handler_thandler,

void *data, intpoll_in, int poll_out);

Int (*event_select_on) (structevent_pool *event_pool, int fd, int idx,

int Poll_in,int poll_out);

Int (*event_unregister) (structevent_pool *event_pool, int fd, int idx);

Int (*event_unregister_close) (structevent_pool *event_pool, int fd,

int idx);

Int (*event_dispatch) (Structevent_pool *event_pool);

Int (*event_reconfigure_threads) (struct Event_pool *event_pool,

Intnewcount);

Int (*event_pool_destroy) (Structevent_pool *event_pool);

};

//event-epoll.c

struct Event_opsevent_ops_epoll = {

. New = Event_pool_new_epoll,

. Event_register = Event_register_epoll,

. event_select_on =event_select_on_epoll,

. Event_unregister = Event_unregister_epoll,

. Event_unregister_close = Event_unregister_close_epoll,

. Event_dispatch = Event_dispatch_epoll,

. Event_reconfigure_threads =event_reconfigure_threads_epoll,

. Event_pool_destroy = Event_pool_destroy_epoll

};

Regardless of that pattern, use epoll and distribute events in Event-dispatch

EVENT_POOL->OPS->EVENT_DISPACTH here is the invocation of a function pointer, referring to Structops's true point is event_dispatch_epoll, and using a separate thread worker to handle all events

Loading of the Xlator

By default, Glusterd uses/etc/glusterfs/glusterd.vol to initialize the Xlator

1:volume Management

2:type Mgmt/glusterd

3:optionrpc-auth.auth-glusterfs on

4:option Rpc-auth.auth-unixon

5:option Rpc-auth.auth-nullon

6:optionrpc-auth-allow-insecure on

7:optiontransport.socket.listen-backlog 128

8:option Event-threads 1

9:option Ping-timeout 0

10:option Transport.socket.read-fail-log off

11:optiontransport.socket.keepalive-interval 2

12:option Transport.socket.keepalive-time 10

13:option Transport-type RDMA

14:option Working-directory/var/lib/glusterd

15:end-volume

Where the XL point is the current xlaotor (MGMT), Glusterd.c:init ()

The RPC service was initialized in Glusterd.c:init ()

A stack structure consisting of different xlator, called a graph