Changeset 5f413b75efe299d3b88314bbcc3c21cbbcec36a7

Timestamp:

02/02/08 04:08:49 (5 years ago)

Author:

Theo Schlossnagle <jesus@omniti.com>

git-committer:

Theo Schlossnagle <jesus@omniti.com> 1201925329 +0000

git-parent:

[78402cdb08d09655ada4b082b6ea4fda7553dadf]

git-author:

Theo Schlossnagle <jesus@omniti.com> 1201925329 +0000

Message:

more work... framework for submitting arbitrary metrics

Files:

• src/.gdbinit (added)
• src/modules/http.c (modified) (11 diffs)
• src/modules/ping_icmp.c (modified) (3 diffs)
• src/noit_conf.c (modified) (2 diffs)
• src/noit_module.h (modified) (1 diff)
• src/noit_poller.c (modified) (13 diffs)
• src/noit_poller.h (modified) (5 diffs)
• src/sample.conf (modified) (2 diffs)

src/modules/http.c

@@ -81 +81 @@
-  serf_check_info_t serf;
-  struct timeval xml_doc_time;
+  char *xpathexpr;
-  xmlDocPtr xml_doc;
+  char *resmod;
+  char *resserv;
-} resmon_check_info_t;
-
@@ -99 +102 @@
-static void serf_log_results(noit_module_t *self, noit_check_t check);
-static void resmon_log_results(noit_module_t *self, noit_check_t check);
+static void resmon_part_log_results(noit_module_t *self, noit_check_t check,
+                                    noit_check_t parent);
-
-static int serf_config(noit_module_t *self, noit_hash_table *options) {
@@ -132 +137 @@
-  char human_buffer[256], code[4], rt[14];
-
+  memset(&current, 0, sizeof(current));
+
-  if(noit_hash_retrieve(check->config, "code", strlen("code"),
-                        (void **)&code_str))
@@ -148 +155 @@
-  noitL(nldeb, "http(%s) [%s]\n", check->target, human_buffer);
-
+  memcpy(&current.whence, &ci->finish_time, sizeof(current.whence));
-  current.duration = duration.tv_sec * 1000 + duration.tv_usec / 1000;
-  current.available = (ci->timed_out || !ci->status.code) ? NP_UNAVAILABLE : NP_AVAILABLE;
-  current.state = (ci->status.code != 200) ? NP_BAD : NP_GOOD;
-  current.status = human_buffer;
-
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
-}
-static void resmon_log_results(noit_module_t *self, noit_check_t check) {
@@ -163 +229 @@
-  xmlDocPtr resmon_results = NULL;
-  xmlXPathContextPtr xpath_ctxt = NULL;
+  xmlXPathObjectPtr pobj = NULL;
+
+  memset(&current, 0, sizeof(current));
-
-  if(ci->body.b) resmon_results = xmlParseMemory(ci->body.b, ci->body.l);
-  if(resmon_results) {
-    xmlXPathObjectPtr pobj;
-    xpath_ctxt = xmlXPathNewContext(resmon_results);
-    pobj = xmlXPathEval((xmlChar *)"/ResmonResults/ResmonResult", xpath_ctxt);
@@ -172 +240 @@
-      if(pobj->type == XPATH_NODESET)
-        services = xmlXPathNodeSetGetLength(pobj->nodesetval);
-    xmlXPathFreeObject(pobj);
-  } else {
-    noitL(nlerr, "Error in resmon doc: %s\n", ci->body.b);
-  }
-
-
-/* Save out results for future dependent checks */ 
+
+
+memcpy(&current.whence, &ci->finish_time, sizeof(current.whence));
-  memcpy(&rci->xml_doc_time, &ci->finish_time, sizeof(ci->finish_time));
-  if(rci->xml_doc) xmlFreeDoc(rci->xml_doc);
-  rci->xml_doc = resmon_results;
-
+  if(rci->xpathexpr) {
+    /* This is actually a part check... we had to do all the work as
+     * it isn't being used as a causal firing from a generic resmon check
+     */
+    resmon_part_log_results_xml(self, check, rci->xml_doc);
+    goto out;
+  }
+
-  sub_timeval(ci->finish_time, check->last_fire_time, &duration);
-
-  snprintf(rt, sizeof(rt), "%.3fms",
-           (float)duration.tv_sec + (float)duration.tv_usec / 1000000.0);
@@ -198 +272 @@
-  current.state = services ? NP_GOOD : NP_BAD;
-  current.status = human_buffer;
-
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
-}
-static int serf_complete(eventer_t e, int mask,
@@ -641 +763 @@
-}
-static int serf_initiate_check(noit_module_t *self, noit_check_t check,
-
+, noit_check_t cause
-  if(!check->closure) check->closure = calloc(1, sizeof(serf_check_info_t));
-  if(once) {
@@ -653 +775 @@
-}
-static int resmon_initiate_check(noit_module_t *self, noit_check_t check,
-
+, noit_check_t parent
-  /* resmon_check_info_t gives us a bit more space */
-  if(!check->closure) check->closure = calloc(1, sizeof(resmon_check_info_t));
@@ -664 +786 @@
-  return 0;
-}
+
+static int resmon_part_initiate_check(noit_module_t *self, noit_check_t check,
+                                      int once, noit_check_t parent) {
+  char xpathexpr[1024];
+  const char *resmod, *resserv;
+  resmon_check_info_t *rci;
+
+  if(!check->closure) check->closure = calloc(1, sizeof(resmon_check_info_t));
+  rci = check->closure;
+  if(!rci->xpathexpr) {
+    if(!noit_hash_retrieve(check->config,
+                           "resmon_module", strlen("resmon_module"),
+                           (void **)&resmod)) {
+      resmod = "DUMMY_MODULE";
+    }
+    if(!noit_hash_retrieve(check->config,
+                           "resmon_service", strlen("resmon_service"),
+                           (void **)&resserv)) {
+      resserv = "DUMMY_SERVICE";
+    }
+    snprintf(xpathexpr, sizeof(xpathexpr),
+             "//ResmonResult[@module=\"%s\" and @service=\"%s\"]/*",
+             resmod, resserv);
+    rci->xpathexpr = strdup(xpathexpr);
+    rci->resmod = strdup(resmod);
+    rci->resserv = strdup(resserv);
+  }
+
+  if(parent && !strcmp(parent->module, "resmon")) {
+    /* Content is cached in the parent */
+    resmon_part_log_results(self, check, parent);
+    return 0;
+  }
+  if(once) {
+    serf_initiate(self, check);
+    return 0;
+  }
+  if(!check->fire_event)
+    serf_schedule_next(self, NULL, check, NULL);
+  return 0;
+}
+
-
-
@@ -702 +866 @@
-};
-
+noit_module_t resmon_part = {
+  NOIT_MODULE_MAGIC,
+  NOIT_MODULE_ABI_VERSION,
+  "resmon_part",
+  "resmon part resource checker",
+  serf_onload,
+  resmon_config,
+  serf_init,
+  resmon_part_initiate_check
+};
+

src/modules/ping_icmp.c

@@ -79 +79 @@
-  struct timeval duration;
-
+  memset(&current, 0, sizeof(current));
+
-  data = (struct check_info *)check->closure;
-  for(i=0; i<data->expected_count; i++) {
@@ -107 +109 @@
-  current.state = (avail < 1.0) ? NP_BAD : NP_GOOD;
-  current.status = human_buffer;
-
+
+
+
+
+
+
+
-}
-static int ping_icmp_timeout(eventer_t e, int mask,
@@ -452 +460 @@
-}
-static int ping_icmp_initiate_check(noit_module_t *self, noit_check_t check,
-
+, noit_check_t cause
-  if(!check->closure) check->closure = calloc(1, sizeof(struct check_info));
-  if(once) {

src/noit_conf.c

@@ -14 +14 @@
-#include "noit_conf.h"
-#include "utils/noit_hash.h"
+#include "utils/noit_log.h"
-
-/* tmp hash impl, replace this with something nice */
@@ -89 +90 @@
-    char *value;
-    node = xmlXPathNodeSetItem(pobj->nodesetval, i);
-SetToString(pobj->nodesetval
+ToString(node
-    noit_hash_replace(table,
-                      strdup((char *)node->name), strlen((char *)node->name),

src/noit_module.h

@@ -22 +22 @@
-  int (*config)(struct _noit_module *, noit_hash_table *options);
-  int (*init)(struct _noit_module *);
-
+
+
-  void *opaque_handle;
-} noit_module_t;

src/noit_poller.c

@@ -26 +26 @@
-  uuid_t foo;
-};
+
-#define UUID_SIZE sizeof(struct uuid_dummy)
-
@@ -65 +66 @@
-    char module[256];
-    char name[256];
+    char oncheck[1024];
+    int no_period = 0;
+    int no_oncheck = 0;
-    int period = 0, timeout = 0;
-    uuid_t uuid, out_uuid;
@@ -88 +92 @@
-      if(!noit_conf_get_stringbuf(sec[i], "../module", module, sizeof(module))) {
-        noitL(noit_stderr, "check uuid: '%s' has no module\n",
-   
+
-        continue;
-      }
@@ -97 +101 @@
-    if(!noit_conf_get_int(sec[i], "period", &period)) {
-      if(!noit_conf_get_int(sec[i], "../period", &period)) {
-
-
-
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
-    }
-    if(!noit_conf_get_int(sec[i], "timeout", &timeout)) {
@@ -107 +126 @@
-      }
-    }
-
+!no_period && 
-      noitL(noit_stderr, "check uuid: '%s' timeout > period\n", uuid_str);
-      timeout = period/2;
@@ -113 +132 @@
-    options = noit_conf_get_hash(sec[i], "config/*");
-    noit_poller_schedule(target, module, name, options,
-
+
+
-    noitL(noit_debug, "loaded uuid: %s\n", uuid_str);
-  }
@@ -129 +149 @@
-    mod = noit_module_lookup(check->module);
-    if(mod) {
-
+
+
-    }
-    else {
-      noitL(noit_stderr, "Cannot find module '%s'\n", check->module);
-
-
-
-
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
-void
-noit_poller_init() {
@@ -143 +203 @@
-                            __check_name_compare);
-  noit_poller_load_checks();
+  noit_poller_make_causal_map();
-  noit_poller_initiate();
-}
@@ -153 +214 @@
-                     u_int32_t period,
-                     u_int32_t timeout,
+                     const char *oncheck,
-                     uuid_t in,
-                     uuid_t out) {
@@ -182 +244 @@
-  new_check->target = strdup(target);
-  new_check->module = strdup(module);
-?strdup(name):
+ ? strdup(name): 
-
-  if(config != NULL) {
@@ -194 +256 @@
-    }
-  }
+  new_check->oncheck = oncheck ? strdup(oncheck) : NULL;
-  new_check->period = period;
-  new_check->timeout = timeout;
@@ -257 +320 @@
-  tmp_check->target = target;
-  tmp_check->name = name;
-&
+
-  free(tmp_check);
-  return check;
-}
-
-
-
-
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
-  int report_change = 0;
+  dep_list_t *dep;
-  if(check->stats.previous.status)
-    free(check->stats.previous.status);
+  noit_hash_destroy(&check->stats.previous.metrics, NULL, __free_metric);
-  memcpy(&check->stats.previous, &check->stats.current, sizeof(stats_t));
-  memcpy(&check->stats.current, newstate, sizeof(stats_t));
@@ -291 +402 @@
-          __noit_check_state_string(check->stats.current.state));
-  }
-
+
+
+
+
+
+
+
+
+
+

src/noit_poller.h

@@ -43 +43 @@
-
-typedef struct {
+  char *metric_name;
+  enum { METRIC_INT, METRIC_FLOAT, METRIC_STRING } metric_type;
+  union {
+    float *f;
+    int *i;
+    char *s;
+  } metric_value;
+} metric_t;
+
+typedef struct {
-  struct timeval whence;
-  int16_t available;
@@ -48 +58 @@
-  u_int32_t duration;
-  char *status;
+  noit_hash_table metrics;
-} stats_t;
-
-
+
+
+
+
+
+
-  uuid_t checkid;
-  int8_t target_family;
@@ -61 +77 @@
-  char *name;
-  noit_hash_table *config;
+  char *oncheck;               /* target`name of the check that fires us */
-  u_int32_t period;            /* period of checks in milliseconds */
-  u_int32_t timeout;           /* timeout of check in milliseconds */
-  u_int32_t flags;             /* NP_KILLED, NP_RUNNING */
-
+  dep_list_t *causal_checks;
-  eventer_t fire_event;
-  struct timeval last_fire_time;
@@ -85 +103 @@
-                       u_int32_t period,
-                       u_int32_t timeout,
+                       const char *oncheck,
-                       uuid_t in,
-                       uuid_t out);
@@ -97 +116 @@
-  noit_poller_lookup_by_name(char *target, char *name);
-
+struct _noit_module;
-API_EXPORT(void)
-
+
+
+
+
+
+
+
+
+
-
-#endif

src/sample.conf

@@ -17 +17 @@
-      <image>http</image>
-      <name>resmon</name>
-  <config
-    <key1>value2</key1
-  <key2>value2</key2
-/config
+</module
+<module
+<image>http</image
+name>resmon_part</name
-    </module>
-    <module>
-      <image>http</image>
-      <name>http</name>
-      <config>
-        <key1>value2</key1>
-        <key2>value2</key2>
-      </config>
-    </module>
-  </modules>
@@ -72 +68 @@
-      </check>
-    </checkgroup>
+    <checkgroup>
+      <module>resmon_part</module>
+      <timeout>1000</timeout>
+      <oncheck>resmon</oncheck>
+      <check uuid="1b4e28ba-2fa1-11d2-883f-b9a761bde3fe">
+        <target>10.225.209.36</target>
+        <config>
+          <resmon_module>FAULTS</resmon_module>
+          <resmon_service>hardware</resmon_service>
+        </config>
+      </check>
+    </checkgroup>
-  </checks>
-</noit>