wdgzyjx/myxce
1
0
Fork 0
forked from lei.chen/myxce
Code Pull requests Activity

save works

Browse source
This commit is contained in:
Rongsong Shen 2026-02-28 17:55:50 +08:00
parent b1faeb94b6
commit b748e30d20
35 changed files with 1247 additions and 2149 deletions
Download patch file Download diff file Expand all files Collapse all files

145
src/controller/worker.c
View file

@ -2,9 +2,11 @@
#include <stdlib.h>
#include "cluster.h"
#include "wobj.h"
#include "worker.h"
#include "xce/cluster_def.h"
#include "xce/errnox.h"
#include "xstring.h"
static int hostmatched(const char* pattern, const char* name)
{
@ -68,7 +70,7 @@ static int update_val_res(cluster_t* c, worker_t* w, xce_resource_t* r,
array = &(c->sharedinst);
} else {
/* update host based resource */
array = &(w->host.resvals);
array = &(w->host->resvals);
}
if (idx < array->caps) {
@ -95,11 +97,14 @@ static int update_val_res(cluster_t* c, worker_t* w, xce_resource_t* r,
static int update_resval(cluster_t* c, worker_t* w, xce_resval_t* val)
{
xce_resource_t* r;
xce_resource_t* r = NULL;
xce_val_t* v = NULL;
int rc;
r = resmgr_lockres(&(c->resmgr), val->name, OBJREF_READ);
rc = key_find(c->resmgr, (void*)val->name, (void**)&r);
if (rc < 0) {
return rc;
}
if (r == NULL) {
ERROR("resource `%s' can not be found", val->name);
return -errno;
@ -114,8 +119,6 @@ static int update_resval(cluster_t* c, worker_t* w, xce_resval_t* val)
rc = update_val_res(c, w, r, v);
obj_deref(r);
return rc;
}
@ -151,14 +154,18 @@ int update_resval_list(cluster_t* c, worker_t* w, xce_resval_list_t* l)
static int mark_host_res(cluster_t* c, worker_t* w, xce_config_t* config)
{
xce_resource_t* r;
xce_resource_t* r = NULL;
char* n = config->name;
int rc;
/* clang-format-off */
xce_val_t val;
/* clang-format-on */
r = resmgr_lockres(&(c->resmgr), n, OBJREF_READ);
rc = key_find(c->resmgr, (void*)n, (void**)&r);
if (rc < 0) {
return rc;
}
if (r == NULL) {
ERROR("resource `%s' can not be found", n);
return -ENOENT;
@ -199,7 +206,6 @@ static int mark_host_res(cluster_t* c, worker_t* w, xce_config_t* config)
}
rc = update_val_res(c, w, r, &val);
obj_deref(r);
return rc;
}
@ -207,16 +213,17 @@ static int add_hostattrs_resvals(cluster_t* c, worker_t* w)
{
#define N_HOST_ATTRS (8)
xce_val_t values[N_HOST_ATTRS] = {
{ .which = Value_u_svalue, .u = { .svalue = w->host.attrs->hosttype } },
{ .which = Value_u_svalue,
.u = { .svalue = w->host.attrs->hostmodel } },
.u = { .svalue = w->host->attrs->hosttype } },
{ .which = Value_u_svalue,
.u = { .svalue = w->host->attrs->hostmodel } },
{ .which = Value_u_fvalue,
.u = { .fvalue = w->host.attrs->cpufactor } },
{ .which = Value_u_fvalue, .u = { .fvalue = w->host.attrs->maxcpus } },
{ .which = Value_u_fvalue, .u = { .fvalue = w->host.attrs->maxmem } },
{ .which = Value_u_fvalue, .u = { .fvalue = w->host.attrs->maxswap } },
{ .which = Value_u_fvalue, .u = { .fvalue = w->host.attrs->maxtmp } },
{ .which = Value_u_fvalue, .u = { .fvalue = w->host.attrs->ndisks } }
.u = { .fvalue = w->host->attrs->cpufactor } },
{ .which = Value_u_fvalue, .u = { .fvalue = w->host->attrs->maxcpus } },
{ .which = Value_u_fvalue, .u = { .fvalue = w->host->attrs->maxmem } },
{ .which = Value_u_fvalue, .u = { .fvalue = w->host->attrs->maxswap } },
{ .which = Value_u_fvalue, .u = { .fvalue = w->host->attrs->maxtmp } },
{ .which = Value_u_fvalue, .u = { .fvalue = w->host->attrs->ndisks } }
};
xce_resval_t resvals[N_HOST_ATTRS] = {
{ .name = "type", .val = &values[0] },
@ -259,7 +266,7 @@ int add_worker(void* cls, endpoint_t* ep, const char* host, hostattrs_t* attrs)
cluster_t* c = (cluster_t*)cls;
/* clang-format off */
worker_t worker = {
.host = {
.host = &(xce_worker_t){
.name = (char*)host,
}
};
@ -273,35 +280,38 @@ int add_worker(void* cls, endpoint_t* ep, const char* host, hostattrs_t* attrs)
return -EINVAL;
}
w = calloc(1, sizeof(*w));
if (w == NULL) {
return -ENOMEM;
}
hconfig = find_matched_hostconfig(c, host);
if (hconfig == NULL) {
return -errno;
}
worker.host.attrs = attrs;
worker.host->attrs = attrs;
w = create_obj_by_typename(&(c->objmgr), "xworker", &worker);
if (w == NULL) {
w->host = makeWorker(worker.host);
if (w->host == NULL) {
return -errno;
}
/* we should hold the update lock */
/* If admin makes the host in locked or closed, we need keep this */
w->host.status &= ~XCE_WORKER_UNAVAIL; /* status OK */
w->host->status &= ~XCE_WORKER_UNAVAIL; /* status OK */
w->ep = ep;
/* override host type? */
if (strcmp(hconfig->hosttype, "!") != 0) {
obj_deref(w->host.attrs->hosttype);
w->host.attrs->hosttype = create_obj_by_typename(
&(c->objmgr), "xstring", hconfig->hosttype);
dereference(w->host->attrs->hosttype);
w->host->attrs->hosttype = makeString(hconfig->hosttype);
}
/* override host model */
if (strcmp(hconfig->hostmodel, "!") != 0) {
obj_deref(w->host.attrs->hostmodel);
w->host.attrs->hostmodel = create_obj_by_typename(
&(c->objmgr), "xstring", hconfig->hostmodel);
dereference(w->host->attrs->hostmodel);
w->host->attrs->hostmodel = makeString(hconfig->hostmodel);
}
add_hostattrs_resvals(c, w);
idx = w->idx;
@ -309,31 +319,31 @@ int add_worker(void* cls, endpoint_t* ep, const char* host, hostattrs_t* attrs)
/* using auto detected or pre configured MXJ? */
nslots = hconfig->mxj;
if (nslots < 0) {
nslots = w->host.attrs->maxcpus;
nslots = w->host->attrs->maxcpus;
}
set_host_slots(c, w, nslots);
if (w->host.dispatch_window_config != NULL) {
free_timewindow_config(w->host.dispatch_window_config);
w->host.dispatch_window_config = NULL;
if (w->host->dispatch_window_config != NULL) {
free_timewindow_config(w->host->dispatch_window_config);
w->host->dispatch_window_config = NULL;
}
if (hconfig->dispatchwindow != NULL) {
w->host.dispatch_window_config
w->host->dispatch_window_config
= dup_timewindow_config(hconfig->dispatchwindow);
}
if (w->host.thresholds != NULL) {
free_resource_thresholds(w->host.thresholds, w->host.nthresholds);
w->host.nthresholds = 0;
w->host.thresholds = NULL;
if (w->host->thresholds != NULL) {
free_resource_thresholds(w->host->thresholds, w->host->nthresholds);
w->host->nthresholds = 0;
w->host->thresholds = NULL;
}
if (hconfig->thresholds != NULL) {
w->host.thresholds = dup_resource_thresholds(hconfig->thresholds,
hconfig->nthresholds);
w->host.nthresholds = hconfig->nthresholds;
w->host->thresholds = dup_resource_thresholds(hconfig->thresholds,
hconfig->nthresholds);
w->host->nthresholds = hconfig->nthresholds;
}
/* add bool resources to host */
@ -354,20 +364,24 @@ int restore_worker(void* cls, xce_worker_t host)
/* do work in add_worker() doing for normal worker join */
/* clang-format off */
worker_t worker = {
.host = {
.host = &(xce_worker_t){
.name = (char*)host.name,
.status = host.status,
},
};
/* clang-format on */
w = create_obj_by_typename(&(c->objmgr), "xworker", &worker);
w = calloc(1, sizeof(*w));
if (w == NULL) {
return -ENOMEM;
}
w->host = makeWorker(worker.host);
if (w->host == NULL) {
return -errno;
}
add_hostattrs_resvals(c, w);
idx = w->idx;
w->host.status |= XCE_WORKER_UNAVAIL;
w->host->status |= XCE_WORKER_UNAVAIL;
DEREF_OBJ(w);
@ -432,12 +446,17 @@ int on_worker_join(void* cls, endpoint_t* ep)
int get_resval(cluster_t* c, char* resname, char* host, xce_val_t* val)
{
xce_resource_t* r;
xce_resource_t* r = NULL;
zarray_t* array = NULL;
xce_val_t* v = NULL;
worker_t* w = NULL;
int rc = 0;
rc = key_find(c->resmgr, (void*)resname, (void**)&r);
if (rc < 0) {
return rc;
}
r = resmgr_lockres(&(c->resmgr), resname, OBJREF_READ);
if (r == NULL) {
return -ENOENT;
}
@ -449,13 +468,15 @@ int get_resval(cluster_t* c, char* resname, char* host, xce_val_t* val)
return -EINVAL;
}
w = workermgr_lock(&(c->workermgr), host, OBJREF_READ);
rc = key_find(c->workermgr, (void*)host, (void**)&w);
if (rc < 0) {
return rc;
}
if (w == NULL) {
DEREF_OBJ(r);
return -ENOENT;
}
array = &(w->host.resvals);
array = &(w->host->resvals);
} else {
array = &(c->sharedinst);
}
@ -464,15 +485,11 @@ int get_resval(cluster_t* c, char* resname, char* host, xce_val_t* val)
v = (xce_val_t*)array->data[r->idx];
}
obj_deref(r);
if (v == NULL) {
DEREF_OBJ(w);
return -ENOENT;
}
memcpy(val, v, sizeof(*val));
DEREF_OBJ(w);
return 1;
}
@ -480,7 +497,7 @@ int mark_worker_unavail(cluster_t* c, worker_t* w)
{
xce_run_hooks(&(c->hooksys), XCE_CONTROLLER_HOOK_WORKER, HOOK_BEFORE_EVENT,
HOOK_DETAIL_WORKER_DEAD, c, w);
w->host.status |= XCE_WORKER_UNAVAIL;
w->host->status |= XCE_WORKER_UNAVAIL;
xce_run_hooks(&(c->hooksys), XCE_CONTROLLER_HOOK_WORKER, HOOK_AFTER_EVENT,
HOOK_DETAIL_WORKER_DEAD, c, w);
@ -491,7 +508,7 @@ int mark_worker_back(cluster_t* c, worker_t* w)
{
xce_run_hooks(&(c->hooksys), XCE_CONTROLLER_HOOK_WORKER, HOOK_BEFORE_EVENT,
HOOK_DETAIL_WORKER_UPDATE, c, w);
w->host.status &= ~XCE_WORKER_UNAVAIL;
w->host->status &= ~XCE_WORKER_UNAVAIL;
xce_run_hooks(&(c->hooksys), XCE_CONTROLLER_HOOK_WORKER, HOOK_AFTER_EVENT,
HOOK_DETAIL_WORKER_UPDATE, c, w);
@ -501,6 +518,7 @@ int mark_worker_back(cluster_t* c, worker_t* w)
int handle_worker_leave(cluster_t* c, int idx)
{
worker_t* w = NULL;
int rc;
TRACE("%s worker %d broken connection", __func__, idx);
@ -514,7 +532,11 @@ int handle_worker_leave(cluster_t* c, int idx)
return -errno;
}
w = workermgr_lock_byid(&(c->workermgr), idx, OBJREF_UPDATE);
rc = vget(c->workermgr, idx, (void**)&w);
if (rc < 0) {
return rc;
}
if (w == NULL) {
ERROR("fail to find worker by id %d", idx);
SET_ERROR(E_XCE_BAD_WORKER);
@ -539,7 +561,12 @@ int worker_ctrl(cluster_t* c, xce_hostctrl_t* ctrl)
return -EINVAL;
}
TRACE("%s worker_ctrl worker %s", __func__, ctrl->hostname);
w = workermgr_lock(&(c->workermgr), ctrl->hostname, OBJREF_UPDATE);
rc = key_find(c->workermgr, (void*)ctrl->hostname, (void**)&w);
if (rc < 0) {
return rc;
}
if (w == NULL) {
PERROR("workermgr_lock");
return -ENOENT;
@ -547,16 +574,16 @@ int worker_ctrl(cluster_t* c, xce_hostctrl_t* ctrl)
switch (ctrl->action) {
case HOST_OPEN:
if (!(w->host.status & XCE_WORKER_CLOSED)) {
if (!(w->host->status & XCE_WORKER_CLOSED)) {
goto clean;
}
w->host.status &= ~XCE_WORKER_CLOSED;
w->host->status &= ~XCE_WORKER_CLOSED;
break;
case HOST_CLOSE:
if (w->host.status & XCE_WORKER_CLOSED) {
if (w->host->status & XCE_WORKER_CLOSED) {
goto clean;
}
w->host.status |= XCE_WORKER_CLOSED;
w->host->status |= XCE_WORKER_CLOSED;
break;
default:
rc = -1;