2 * collectd - src/ceph.c
3 * Copyright (C) 2011 New Dream Network
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License as published by the
7 * Free Software Foundation; only version 2 of the License is applicable.
9 * This program is distributed in the hope that it will be useful, but
10 * WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * General Public License for more details.
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 * Colin McCabe <cmccabe@alumni.cmu.edu>
20 * Dennis Zou <yunzou@cisco.com>
21 * Dan Ryder <daryder@cisco.com>
30 #include <arpa/inet.h>
33 #include <yajl/yajl_parse.h>
34 #if HAVE_YAJL_YAJL_VERSION_H
35 #include <yajl/yajl_version.h>
45 #include <sys/socket.h>
47 #include <sys/types.h>
53 #define RETRY_AVGCOUNT -1
55 #if defined(YAJL_MAJOR) && (YAJL_MAJOR > 1)
56 # define HAVE_YAJL_V2 1
59 #define RETRY_ON_EINTR(ret, expr) \
69 /** Timeout interval in seconds */
70 #define CEPH_TIMEOUT_INTERVAL 1
72 /** Maximum path length for a UNIX domain socket on this system */
73 #define UNIX_DOMAIN_SOCK_PATH_MAX (sizeof(((struct sockaddr_un*)0)->sun_path))
75 /** Yajl callback returns */
76 #define CEPH_CB_CONTINUE 1
77 #define CEPH_CB_ABORT 0
80 typedef size_t yajl_len_t;
82 typedef unsigned int yajl_len_t;
85 /** Number of types for ceph defined in types.db */
86 #define CEPH_DSET_TYPES_NUM 3
87 /** ceph types enum */
93 DSET_TYPE_UNFOUND = 1000
96 /** Valid types for ceph defined in types.db */
97 const char * ceph_dset_types [CEPH_DSET_TYPES_NUM] =
98 {"ceph_latency", "ceph_bytes", "ceph_rate"};
100 /******* ceph_daemon *******/
103 /** Version of the admin_socket interface */
106 char name[DATA_MAX_NAME_LEN];
108 /** Path to the socket that we use to talk to the ceph daemon */
109 char asok_path[UNIX_DOMAIN_SOCK_PATH_MAX];
111 /** Number of counters */
113 /** Track ds types */
115 /** Track ds names to match with types */
119 /******* JSON parsing *******/
120 typedef int (*node_handler_t)(void *, const char*, const char*);
122 /** Track state and handler while parsing JSON */
125 node_handler_t handler;
128 char key[DATA_MAX_NAME_LEN];
130 } state[YAJL_MAX_DEPTH];
133 typedef struct yajl_struct yajl_struct;
136 * Keep track of last data for latency values so we can calculate rate
139 struct last_data **last_poll_data = NULL;
140 /** index of last poll data */
143 enum perfcounter_type_d
145 PERFCOUNTER_LATENCY = 0x4, PERFCOUNTER_DERIVE = 0x8,
148 /** Give user option to use default (long run = since daemon started) avg */
149 static int long_run_latency_avg = 0;
152 * Give user option to use default type for special cases -
153 * filestore.journal_wr_bytes is currently only metric here. Ceph reports the
154 * type as a sum/count pair and will calculate it the same as a latency value.
155 * All other "bytes" metrics (excluding the used/capacity bytes for the OSD)
156 * use the DERIVE type. Unless user specifies to use given type, convert this
157 * metric to use DERIVE.
159 static int convert_special_metrics = 1;
161 /** Array of daemons to monitor */
162 static struct ceph_daemon **g_daemons = NULL;
164 /** Number of elements in g_daemons */
165 static int g_num_daemons = 0;
168 * A set of data that we build up in memory while parsing the JSON.
172 /** ceph daemon we are processing data for*/
173 struct ceph_daemon *d;
174 /** track avgcount across counters for avgcount/sum latency pairs */
176 /** current index of counters - used to get type of counter */
178 /** do we already have an avgcount for latency pair */
181 * similar to index, but current index of latency type counters -
182 * used to get last poll data of counter
186 * values list - maintain across counters since
187 * host/plugin/plugin instance are always the same
193 * A set of count/sum pairs to keep track of latency types and get difference
194 * between this poll data and last poll data.
198 char ds_name[DATA_MAX_NAME_LEN];
204 /******* network I/O *******/
207 CSTATE_UNCONNECTED = 0,
208 CSTATE_WRITE_REQUEST,
216 ASOK_REQ_VERSION = 0,
219 ASOK_REQ_NONE = 1000,
224 /** The Ceph daemon that we're talking to */
225 struct ceph_daemon *d;
228 uint32_t request_type;
230 /** The connection state */
233 /** The socket we use to talk to this daemon */
236 /** The amount of data remaining to read / write. */
239 /** Length of the JSON to read */
242 /** Buffer containing JSON data */
245 /** Keep data important to yajl processing */
246 struct yajl_struct yajl;
249 static int ceph_cb_null(void *ctx)
251 return CEPH_CB_CONTINUE;
254 static int ceph_cb_boolean(void *ctx, int bool_val)
256 return CEPH_CB_CONTINUE;
260 ceph_cb_number(void *ctx, const char *number_val, yajl_len_t number_len)
262 yajl_struct *yajl = (yajl_struct*)ctx;
263 char buffer[number_len+1];
264 int i, latency_type = 0, result;
267 memcpy(buffer, number_val, number_len);
268 buffer[sizeof(buffer) - 1] = 0;
270 ssnprintf(key, yajl->state[0].key_len, "%s", yajl->state[0].key);
271 for(i = 1; i < yajl->depth; i++)
273 if((i == yajl->depth-1) && ((strcmp(yajl->state[i].key,"avgcount") == 0)
274 || (strcmp(yajl->state[i].key,"sum") == 0)))
276 if(convert_special_metrics)
279 * Special case for filestore:JournalWrBytes. For some reason,
280 * Ceph schema encodes this as a count/sum pair while all
281 * other "Bytes" data (excluding used/capacity bytes for OSD
282 * space) uses a single "Derive" type. To spare further
283 * confusion, keep this KPI as the same type of other "Bytes".
284 * Instead of keeping an "average" or "rate", use the "sum" in
285 * the pair and assign that to the derive value.
287 if((strcmp(yajl->state[i-1].key, "journal_wr_bytes") == 0) &&
288 (strcmp(yajl->state[i-2].key,"filestore") == 0) &&
289 (strcmp(yajl->state[i].key,"avgcount") == 0))
291 DEBUG("ceph plugin: Skipping avgcount for filestore.JournalWrBytes");
292 yajl->depth = (yajl->depth - 1);
293 return CEPH_CB_CONTINUE;
296 //probably a avgcount/sum pair. if not - we'll try full key later
300 strncat(key, ".", 1);
301 strncat(key, yajl->state[i].key, yajl->state[i].key_len+1);
304 result = yajl->handler(yajl->handler_arg, buffer, key);
306 if((result == RETRY_AVGCOUNT) && latency_type)
308 strncat(key, ".", 1);
309 strncat(key, yajl->state[yajl->depth-1].key,
310 yajl->state[yajl->depth-1].key_len+1);
311 result = yajl->handler(yajl->handler_arg, buffer, key);
314 if(result == -ENOMEM)
316 ERROR("ceph plugin: memory allocation failed");
317 return CEPH_CB_ABORT;
320 yajl->depth = (yajl->depth - 1);
321 return CEPH_CB_CONTINUE;
324 static int ceph_cb_string(void *ctx, const unsigned char *string_val,
325 yajl_len_t string_len)
327 return CEPH_CB_CONTINUE;
330 static int ceph_cb_start_map(void *ctx)
332 return CEPH_CB_CONTINUE;
336 ceph_cb_map_key(void *ctx, const unsigned char *key, yajl_len_t string_len)
338 yajl_struct *yajl = (yajl_struct*)ctx;
340 if((yajl->depth+1) >= YAJL_MAX_DEPTH)
342 ERROR("ceph plugin: depth exceeds max, aborting.");
343 return CEPH_CB_ABORT;
346 char buffer[string_len+1];
348 memcpy(buffer, key, string_len);
349 buffer[sizeof(buffer) - 1] = 0;
351 snprintf(yajl->state[yajl->depth].key, sizeof(buffer), "%s", buffer);
352 yajl->state[yajl->depth].key_len = sizeof(buffer);
353 yajl->depth = (yajl->depth + 1);
355 return CEPH_CB_CONTINUE;
358 static int ceph_cb_end_map(void *ctx)
360 yajl_struct *yajl = (yajl_struct*)ctx;
362 yajl->depth = (yajl->depth - 1);
363 return CEPH_CB_CONTINUE;
366 static int ceph_cb_start_array(void *ctx)
368 return CEPH_CB_CONTINUE;
371 static int ceph_cb_end_array(void *ctx)
373 return CEPH_CB_CONTINUE;
376 static yajl_callbacks callbacks = {
390 static void ceph_daemon_print(const struct ceph_daemon *d)
392 DEBUG("ceph plugin: name=%s, asok_path=%s", d->name, d->asok_path);
395 static void ceph_daemons_print(void)
398 for(i = 0; i < g_num_daemons; ++i)
400 ceph_daemon_print(g_daemons[i]);
404 static void ceph_daemon_free(struct ceph_daemon *d)
407 for(; i < d->ds_num; i++)
409 sfree(d->ds_names[i]);
417 * Compact ds name by removing special characters and trimming length to
418 * DATA_MAX_NAME_LEN if necessary
420 static void compact_ds_name(char *source, char *dest)
423 char *save_ptr = NULL, *tmp_ptr = source;
426 char tmp[DATA_MAX_NAME_LEN];
427 size_t key_chars_remaining = (DATA_MAX_NAME_LEN-1);
430 memset(tmp, 0, sizeof(tmp));
431 if(source == NULL || dest == NULL || source[0] == '\0' || dest[0] != '\0')
435 size_t src_len = strlen(source);
436 snprintf(len_str, sizeof(len_str), "%zu", src_len);
437 unsigned char append_status = 0x0;
438 append_status |= (source[src_len - 1] == '-') ? 0x1 : 0x0;
439 append_status |= (source[src_len - 1] == '+') ? 0x2 : 0x0;
440 while ((keys[keys_num] = strtok_r(tmp_ptr, ":_-+", &save_ptr)) != NULL)
443 /** capitalize 1st char **/
444 keys[keys_num][0] = toupper(keys[keys_num][0]);
451 /** concatenate each part of source string **/
452 for(i = 0; i < keys_num; i++)
454 strncat(tmp, keys[i], key_chars_remaining);
455 key_chars_remaining -= strlen(keys[i]);
457 tmp[DATA_MAX_NAME_LEN - 1] = '\0';
458 /** to coordinate limitation of length of type_instance
459 * we will truncate ds_name
460 * when the its length is more than
463 if(strlen(tmp) > DATA_MAX_NAME_LEN - 1)
465 append_status |= 0x4;
466 /** we should reserve space for
471 if(append_status & 0x1)
473 /** we should reserve space for
478 if(append_status & 0x2)
480 /** we should reserve space for
485 snprintf(dest, DATA_MAX_NAME_LEN - reserved, "%s", tmp);
486 offset = strlen(dest);
487 switch (append_status)
490 memcpy(dest + offset, "Minus", 5);
493 memcpy(dest + offset, "Plus", 5);
496 memcpy(dest + offset, len_str, 2);
499 memcpy(dest + offset, "Minus", 5);
500 memcpy(dest + offset + 5, len_str, 2);
503 memcpy(dest + offset, "Plus", 4);
504 memcpy(dest + offset + 4, len_str, 2);
512 * Parse key to remove "type" if this is for schema and initiate compaction
514 static int parse_keys(const char *key_str, char *ds_name)
517 size_t ds_name_len = 0;
519 * allow up to 100 characters before compaction - compact_ds_name will not
520 * allow more than DATA_MAX_NAME_LEN chars
522 int max_str_len = 100;
523 char tmp_ds_name[max_str_len];
524 memset(tmp_ds_name, 0, sizeof(tmp_ds_name));
525 if(ds_name == NULL || key_str == NULL || key_str[0] == '\0' ||
530 if((ptr = strchr(key_str, '.')) == NULL
531 || (rptr = strrchr(key_str, '.')) == NULL)
533 memcpy(tmp_ds_name, key_str, max_str_len - 1);
536 ds_name_len = (rptr - ptr) > max_str_len ? max_str_len : (rptr - ptr);
537 if((ds_name_len == 0) || strncmp(rptr + 1, "type", 4))
538 { /** copy whole key **/
539 memcpy(tmp_ds_name, key_str, max_str_len - 1);
542 {/** more than two keys **/
543 memcpy(tmp_ds_name, key_str, ((rptr - key_str) > (max_str_len - 1) ?
544 (max_str_len - 1) : (rptr - key_str)));
547 compact: compact_ds_name(tmp_ds_name, ds_name);
552 * while parsing ceph admin socket schema, save counter name and type for later
555 static int ceph_daemon_add_ds_entry(struct ceph_daemon *d, const char *name,
559 char ds_name[DATA_MAX_NAME_LEN];
560 memset(ds_name, 0, sizeof(ds_name));
562 if(convert_special_metrics)
565 * Special case for filestore:JournalWrBytes. For some reason, Ceph
566 * schema encodes this as a count/sum pair while all other "Bytes" data
567 * (excluding used/capacity bytes for OSD space) uses a single "Derive"
568 * type. To spare further confusion, keep this KPI as the same type of
569 * other "Bytes". Instead of keeping an "average" or "rate", use the
570 * "sum" in the pair and assign that to the derive value.
572 if((strcmp(name,"filestore.journal_wr_bytes.type") == 0))
578 d->ds_names = realloc(d->ds_names, sizeof(char *) * (d->ds_num + 1));
584 d->ds_types = realloc(d->ds_types, sizeof(uint32_t) * (d->ds_num + 1));
590 d->ds_names[d->ds_num] = malloc(sizeof(char) * DATA_MAX_NAME_LEN);
591 if(!d->ds_names[d->ds_num])
596 type = (pc_type & PERFCOUNTER_DERIVE) ? DSET_RATE :
597 ((pc_type & PERFCOUNTER_LATENCY) ? DSET_LATENCY : DSET_BYTES);
598 d->ds_types[d->ds_num] = type;
600 if(parse_keys(name, ds_name))
605 sstrncpy(d->ds_names[d->ds_num], ds_name, DATA_MAX_NAME_LEN -1);
606 d->ds_num = (d->ds_num + 1);
611 /******* ceph_config *******/
612 static int cc_handle_str(struct oconfig_item_s *item, char *dest, int dest_len)
615 if(item->values_num != 1)
619 if(item->values[0].type != OCONFIG_TYPE_STRING)
623 val = item->values[0].value.string;
624 if(snprintf(dest, dest_len, "%s", val) > (dest_len - 1))
626 ERROR("ceph plugin: configuration parameter '%s' is too long.\n",
628 return -ENAMETOOLONG;
633 static int cc_handle_bool(struct oconfig_item_s *item, int *dest)
635 if(item->values_num != 1)
640 if(item->values[0].type != OCONFIG_TYPE_BOOLEAN)
645 *dest = (item->values[0].value.boolean) ? 1 : 0;
649 static int cc_add_daemon_config(oconfig_item_t *ci)
652 struct ceph_daemon *array, *nd, cd;
653 memset(&cd, 0, sizeof(struct ceph_daemon));
655 if((ci->values_num != 1) || (ci->values[0].type != OCONFIG_TYPE_STRING))
657 WARNING("ceph plugin: `Daemon' blocks need exactly one string "
662 ret = cc_handle_str(ci, cd.name, DATA_MAX_NAME_LEN);
668 for(i=0; i < ci->children_num; i++)
670 oconfig_item_t *child = ci->children + i;
672 if(strcasecmp("SocketPath", child->key) == 0)
674 ret = cc_handle_str(child, cd.asok_path, sizeof(cd.asok_path));
682 WARNING("ceph plugin: ignoring unknown option %s", child->key);
685 if(cd.name[0] == '\0')
687 ERROR("ceph plugin: you must configure a daemon name.\n");
690 else if(cd.asok_path[0] == '\0')
692 ERROR("ceph plugin(name=%s): you must configure an administrative "
693 "socket path.\n", cd.name);
696 else if(!((cd.asok_path[0] == '/') ||
697 (cd.asok_path[0] == '.' && cd.asok_path[1] == '/')))
699 ERROR("ceph plugin(name=%s): administrative socket paths must begin "
700 "with '/' or './' Can't parse: '%s'\n", cd.name, cd.asok_path);
703 array = realloc(g_daemons,
704 sizeof(struct ceph_daemon *) * (g_num_daemons + 1));
707 /* The positive return value here indicates that this is a
708 * runtime error, not a configuration error. */
711 g_daemons = (struct ceph_daemon**) array;
712 nd = malloc(sizeof(struct ceph_daemon));
717 memcpy(nd, &cd, sizeof(struct ceph_daemon));
718 g_daemons[g_num_daemons++] = nd;
722 static int ceph_config(oconfig_item_t *ci)
726 for(i = 0; i < ci->children_num; ++i)
728 oconfig_item_t *child = ci->children + i;
729 if(strcasecmp("Daemon", child->key) == 0)
731 ret = cc_add_daemon_config(child);
734 ERROR("ceph plugin: Couldn't allocate memory");
739 //process other daemons and ignore this one
743 else if(strcasecmp("LongRunAvgLatency", child->key) == 0)
745 ret = cc_handle_bool(child, &long_run_latency_avg);
751 else if(strcasecmp("ConvertSpecialMetricTypes", child->key) == 0)
753 ret = cc_handle_bool(child, &convert_special_metrics);
761 WARNING("ceph plugin: ignoring unknown option %s", child->key);
768 * Parse JSON and get error message if present
771 traverse_json(const unsigned char *json, uint32_t json_len, yajl_handle hand)
773 yajl_status status = yajl_parse(hand, json, json_len);
778 case yajl_status_error:
779 msg = yajl_get_error(hand, /* verbose = */ 1,
780 /* jsonText = */ (unsigned char *) json,
781 (unsigned int) json_len);
782 ERROR ("ceph plugin: yajl_parse failed: %s", msg);
783 yajl_free_error(hand, msg);
785 case yajl_status_client_canceled:
793 * Add entry for each counter while parsing schema
796 node_handler_define_schema(void *arg, const char *val, const char *key)
798 struct ceph_daemon *d = (struct ceph_daemon *) arg;
801 DEBUG("ceph plugin: ceph_daemon_add_ds_entry(d=%s,key=%s,pc_type=%04x)",
802 d->name, key, pc_type);
803 return ceph_daemon_add_ds_entry(d, key, pc_type);
807 * Latency counter does not yet have an entry in last poll data - add it.
809 static int add_last(const char *ds_n, double cur_sum, uint64_t cur_count)
811 last_poll_data[last_idx] = malloc(1 * sizeof(struct last_data));
812 if(!last_poll_data[last_idx])
816 sstrncpy(last_poll_data[last_idx]->ds_name,ds_n,
817 sizeof(last_poll_data[last_idx]->ds_name));
818 last_poll_data[last_idx]->last_sum = cur_sum;
819 last_poll_data[last_idx]->last_count = cur_count;
825 * Update latency counter or add new entry if it doesn't exist
827 static int update_last(const char *ds_n, int index, double cur_sum,
830 if((last_idx > index) && (strcmp(last_poll_data[index]->ds_name, ds_n) == 0))
832 last_poll_data[index]->last_sum = cur_sum;
833 last_poll_data[index]->last_count = cur_count;
839 last_poll_data = malloc(1 * sizeof(struct last_data *));
847 struct last_data **tmp_last = realloc(last_poll_data,
848 ((last_idx+1) * sizeof(struct last_data *)));
853 last_poll_data = tmp_last;
855 return add_last(ds_n, cur_sum, cur_count);
859 * Calculate average b/t current data and last poll data
860 * if last poll data exists
862 static double get_last_avg(const char *ds_n, int index,
863 double cur_sum, uint64_t cur_count)
865 double result = -1.1, sum_delt = 0.0;
866 uint64_t count_delt = 0;
867 if((last_idx > index) &&
868 (strcmp(last_poll_data[index]->ds_name, ds_n) == 0) &&
869 (cur_count > last_poll_data[index]->last_count))
871 sum_delt = (cur_sum - last_poll_data[index]->last_sum);
872 count_delt = (cur_count - last_poll_data[index]->last_count);
873 result = (sum_delt / count_delt);
880 if(update_last(ds_n, index, cur_sum, cur_count) == -ENOMEM)
888 * If using index guess failed, resort to searching for counter name
890 static uint32_t backup_search_for_type(struct ceph_daemon *d, char *ds_name)
893 for(; idx < d->ds_num; idx++)
895 if(strcmp(d->ds_names[idx], ds_name) == 0)
897 return d->ds_types[idx];
900 return DSET_TYPE_UNFOUND;
904 * Process counter data and dispatch values
906 static int node_handler_fetch_data(void *arg, const char *val, const char *key)
911 struct values_tmp *vtmp = (struct values_tmp*) arg;
912 uint32_t type = DSET_TYPE_UNFOUND;
913 int index = vtmp->index;
915 char ds_name[DATA_MAX_NAME_LEN];
916 memset(ds_name, 0, sizeof(ds_name));
918 if(parse_keys(key, ds_name))
923 if(index >= vtmp->d->ds_num)
925 //don't overflow bounds of array
926 index = (vtmp->d->ds_num - 1);
930 * counters should remain in same order we parsed schema... we maintain the
931 * index variable to keep track of current point in list of counters. first
932 * use index to guess point in array for retrieving type. if that doesn't
933 * work, use the old way to get the counter type
935 if(strcmp(ds_name, vtmp->d->ds_names[index]) == 0)
938 type = vtmp->d->ds_types[index];
940 else if((index > 0) && (strcmp(ds_name, vtmp->d->ds_names[index-1]) == 0))
943 type = vtmp->d->ds_types[index-1];
946 if(type == DSET_TYPE_UNFOUND)
948 //couldn't find right type by guessing, check the old way
949 type = backup_search_for_type(vtmp->d, ds_name);
955 if(vtmp->avgcount_exists == -1)
957 sscanf(val, "%" PRIu64, &vtmp->avgcount);
958 vtmp->avgcount_exists = 0;
959 //return after saving avgcount - don't dispatch value
960 //until latency calculation
966 sscanf(val, "%lf", &sum);
967 DEBUG("ceph plugin: avgcount:%" PRIu64,vtmp->avgcount);
968 DEBUG("ceph plugin: sum:%lf",sum);
970 if(vtmp->avgcount == 0)
975 /** User wants latency values as long run avg */
976 if(long_run_latency_avg)
978 result = (sum / vtmp->avgcount);
979 DEBUG("ceph plugin: uv->gauge = sumd / avgcounti = :%lf", result);
983 result = get_last_avg(ds_name, vtmp->latency_index, sum, vtmp->avgcount);
984 if(result == -ENOMEM)
988 DEBUG("ceph plugin: uv->gauge = (sumd_now - sumd_last) / "
989 "(avgcounti_now - avgcounti_last) = :%lf", result);
993 vtmp->avgcount_exists = -1;
994 vtmp->latency_index = (vtmp->latency_index + 1);
998 sscanf(val, "%lf", &tmp_d);
1000 DEBUG("ceph plugin: uv->gauge = %lf",uv.gauge);
1003 sscanf(val, "%" PRIu64, &tmp_u);
1005 DEBUG("ceph plugin: uv->derive = %" PRIu64 "",(uint64_t)uv.derive);
1007 case DSET_TYPE_UNFOUND:
1009 ERROR("ceph plugin: ds %s was not properly initialized.", ds_name);
1013 sstrncpy(vtmp->vlist.type, ceph_dset_types[type], sizeof(vtmp->vlist.type));
1014 sstrncpy(vtmp->vlist.type_instance, ds_name, sizeof(vtmp->vlist.type_instance));
1015 vtmp->vlist.values = &uv;
1016 vtmp->vlist.values_len = 1;
1018 DEBUG("ceph plugin: dispatching %s\n", ds_name);
1019 vtmp->index = (vtmp->index + 1);
1020 plugin_dispatch_values(&vtmp->vlist);
1025 static int cconn_connect(struct cconn *io)
1027 struct sockaddr_un address;
1029 if(io->state != CSTATE_UNCONNECTED)
1031 ERROR("ceph plugin: cconn_connect: io->state != CSTATE_UNCONNECTED");
1034 fd = socket(PF_UNIX, SOCK_STREAM, 0);
1038 ERROR("ceph plugin: cconn_connect: socket(PF_UNIX, SOCK_STREAM, 0) "
1039 "failed: error %d", err);
1042 memset(&address, 0, sizeof(struct sockaddr_un));
1043 address.sun_family = AF_UNIX;
1044 snprintf(address.sun_path, sizeof(address.sun_path), "%s",
1047 connect(fd, (struct sockaddr *) &address, sizeof(struct sockaddr_un)));
1050 ERROR("ceph plugin: cconn_connect: connect(%d) failed: error %d",
1055 flags = fcntl(fd, F_GETFL, 0);
1056 if(fcntl(fd, F_SETFL, flags | O_NONBLOCK) != 0)
1059 ERROR("ceph plugin: cconn_connect: fcntl(%d, O_NONBLOCK) error %d",
1064 io->state = CSTATE_WRITE_REQUEST;
1071 static void cconn_close(struct cconn *io)
1073 io->state = CSTATE_UNCONNECTED;
1077 RETRY_ON_EINTR(res, close(io->asok));
1086 /* Process incoming JSON counter data */
1088 cconn_process_data(struct cconn *io, yajl_struct *yajl, yajl_handle hand)
1091 struct values_tmp *vtmp = calloc(1, sizeof(struct values_tmp) * 1);
1097 vtmp->vlist = (value_list_t)VALUE_LIST_INIT;
1098 sstrncpy(vtmp->vlist.host, hostname_g, sizeof(vtmp->vlist.host));
1099 sstrncpy(vtmp->vlist.plugin, "ceph", sizeof(vtmp->vlist.plugin));
1100 sstrncpy(vtmp->vlist.plugin_instance, io->d->name, sizeof(vtmp->vlist.plugin_instance));
1103 vtmp->avgcount_exists = -1;
1104 vtmp->latency_index = 0;
1106 yajl->handler_arg = vtmp;
1107 ret = traverse_json(io->json, io->json_len, hand);
1113 * Initiate JSON parsing and print error if one occurs
1115 static int cconn_process_json(struct cconn *io)
1117 if((io->request_type != ASOK_REQ_DATA) &&
1118 (io->request_type != ASOK_REQ_SCHEMA))
1127 hand = yajl_alloc(&callbacks,
1129 /* alloc funcs = */ NULL,
1131 /* alloc funcs = */ NULL, NULL,
1133 /* context = */ (void *)(&io->yajl));
1137 ERROR ("ceph plugin: yajl_alloc failed.");
1143 switch(io->request_type)
1146 io->yajl.handler = node_handler_fetch_data;
1147 result = cconn_process_data(io, &io->yajl, hand);
1149 case ASOK_REQ_SCHEMA:
1150 io->yajl.handler = node_handler_define_schema;
1151 io->yajl.handler_arg = io->d;
1152 result = traverse_json(io->json, io->json_len, hand);
1162 status = yajl_complete_parse(hand);
1164 status = yajl_parse_complete(hand);
1167 if (status != yajl_status_ok)
1169 unsigned char *errmsg = yajl_get_error (hand, /* verbose = */ 0,
1170 /* jsonText = */ NULL, /* jsonTextLen = */ 0);
1171 ERROR ("ceph plugin: yajl_parse_complete failed: %s",
1173 yajl_free_error (hand, errmsg);
1183 static int cconn_validate_revents(struct cconn *io, int revents)
1185 if(revents & POLLERR)
1187 ERROR("ceph plugin: cconn_validate_revents(name=%s): got POLLERR",
1193 case CSTATE_WRITE_REQUEST:
1194 return (revents & POLLOUT) ? 0 : -EINVAL;
1195 case CSTATE_READ_VERSION:
1196 case CSTATE_READ_AMT:
1197 case CSTATE_READ_JSON:
1198 return (revents & POLLIN) ? 0 : -EINVAL;
1199 return (revents & POLLIN) ? 0 : -EINVAL;
1201 ERROR("ceph plugin: cconn_validate_revents(name=%s) got to "
1202 "illegal state on line %d", io->d->name, __LINE__);
1207 /** Handle a network event for a connection */
1208 static int cconn_handle_event(struct cconn *io)
1213 case CSTATE_UNCONNECTED:
1214 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1215 "state on line %d", io->d->name, __LINE__);
1218 case CSTATE_WRITE_REQUEST:
1221 snprintf(cmd, sizeof(cmd), "%s%d%s", "{ \"prefix\": \"",
1222 io->request_type, "\" }\n");
1223 size_t cmd_len = strlen(cmd);
1225 write(io->asok, ((char*)&cmd) + io->amt, cmd_len - io->amt));
1226 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,amt=%d,ret=%d)",
1227 io->d->name, io->state, io->amt, ret);
1233 if(io->amt >= cmd_len)
1236 switch (io->request_type)
1238 case ASOK_REQ_VERSION:
1239 io->state = CSTATE_READ_VERSION;
1242 io->state = CSTATE_READ_AMT;
1248 case CSTATE_READ_VERSION:
1251 read(io->asok, ((char*)(&io->d->version)) + io->amt,
1252 sizeof(io->d->version) - io->amt));
1253 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1254 io->d->name, io->state, ret);
1260 if(io->amt >= sizeof(io->d->version))
1262 io->d->version = ntohl(io->d->version);
1263 if(io->d->version != 1)
1265 ERROR("ceph plugin: cconn_handle_event(name=%s) not "
1266 "expecting version %d!", io->d->name, io->d->version);
1269 DEBUG("ceph plugin: cconn_handle_event(name=%s): identified as "
1270 "version %d", io->d->name, io->d->version);
1273 io->request_type = ASOK_REQ_SCHEMA;
1277 case CSTATE_READ_AMT:
1280 read(io->asok, ((char*)(&io->json_len)) + io->amt,
1281 sizeof(io->json_len) - io->amt));
1282 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1283 io->d->name, io->state, ret);
1289 if(io->amt >= sizeof(io->json_len))
1291 io->json_len = ntohl(io->json_len);
1293 io->state = CSTATE_READ_JSON;
1294 io->json = calloc(1, io->json_len + 1);
1297 ERROR("ceph plugin: error callocing io->json");
1303 case CSTATE_READ_JSON:
1306 read(io->asok, io->json + io->amt, io->json_len - io->amt));
1307 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1308 io->d->name, io->state, ret);
1314 if(io->amt >= io->json_len)
1316 ret = cconn_process_json(io);
1322 io->request_type = ASOK_REQ_NONE;
1327 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1328 "state on line %d", io->d->name, __LINE__);
1333 static int cconn_prepare(struct cconn *io, struct pollfd* fds)
1336 if(io->request_type == ASOK_REQ_NONE)
1338 /* The request has already been serviced. */
1341 else if((io->request_type == ASOK_REQ_DATA) && (io->d->ds_num == 0))
1343 /* If there are no counters to report on, don't bother
1350 case CSTATE_UNCONNECTED:
1351 ret = cconn_connect(io);
1361 fds->events = POLLOUT;
1363 case CSTATE_WRITE_REQUEST:
1365 fds->events = POLLOUT;
1367 case CSTATE_READ_VERSION:
1368 case CSTATE_READ_AMT:
1369 case CSTATE_READ_JSON:
1371 fds->events = POLLIN;
1374 ERROR("ceph plugin: cconn_prepare(name=%s) got to illegal state "
1375 "on line %d", io->d->name, __LINE__);
1380 /** Returns the difference between two struct timevals in milliseconds.
1381 * On overflow, we return max/min int.
1383 static int milli_diff(const struct timeval *t1, const struct timeval *t2)
1386 int sec_diff = t1->tv_sec - t2->tv_sec;
1387 int usec_diff = t1->tv_usec - t2->tv_usec;
1388 ret = usec_diff / 1000;
1389 ret += (sec_diff * 1000);
1390 return (ret > INT_MAX) ? INT_MAX : ((ret < INT_MIN) ? INT_MIN : (int)ret);
1393 /** This handles the actual network I/O to talk to the Ceph daemons.
1395 static int cconn_main_loop(uint32_t request_type)
1397 int i, ret, some_unreachable = 0;
1398 struct timeval end_tv;
1399 struct cconn io_array[g_num_daemons];
1401 DEBUG("ceph plugin: entering cconn_main_loop(request_type = %d)", request_type);
1403 /* create cconn array */
1404 memset(io_array, 0, sizeof(io_array));
1405 for(i = 0; i < g_num_daemons; ++i)
1407 io_array[i].d = g_daemons[i];
1408 io_array[i].request_type = request_type;
1409 io_array[i].state = CSTATE_UNCONNECTED;
1412 /** Calculate the time at which we should give up */
1413 gettimeofday(&end_tv, NULL);
1414 end_tv.tv_sec += CEPH_TIMEOUT_INTERVAL;
1420 struct cconn *polled_io_array[g_num_daemons];
1421 struct pollfd fds[g_num_daemons];
1422 memset(fds, 0, sizeof(fds));
1424 for(i = 0; i < g_num_daemons; ++i)
1426 struct cconn *io = io_array + i;
1427 ret = cconn_prepare(io, fds + nfds);
1430 WARNING("ceph plugin: cconn_prepare(name=%s,i=%d,st=%d)=%d",
1431 io->d->name, i, io->state, ret);
1433 io->request_type = ASOK_REQ_NONE;
1434 some_unreachable = 1;
1438 DEBUG("ceph plugin: did cconn_prepare(name=%s,i=%d,st=%d)",
1439 io->d->name, i, io->state);
1440 polled_io_array[nfds++] = io_array + i;
1447 DEBUG("ceph plugin: cconn_main_loop: no more cconn to manage.");
1450 gettimeofday(&tv, NULL);
1451 diff = milli_diff(&end_tv, &tv);
1456 WARNING("ceph plugin: cconn_main_loop: timed out.");
1459 RETRY_ON_EINTR(ret, poll(fds, nfds, diff));
1462 ERROR("ceph plugin: poll(2) error: %d", ret);
1465 for(i = 0; i < nfds; ++i)
1467 struct cconn *io = polled_io_array[i];
1468 int revents = fds[i].revents;
1473 else if(cconn_validate_revents(io, revents))
1475 WARNING("ceph plugin: cconn(name=%s,i=%d,st=%d): "
1476 "revents validation error: "
1477 "revents=0x%08x", io->d->name, i, io->state, revents);
1479 io->request_type = ASOK_REQ_NONE;
1480 some_unreachable = 1;
1484 int ret = cconn_handle_event(io);
1487 WARNING("ceph plugin: cconn_handle_event(name=%s,"
1488 "i=%d,st=%d): error %d", io->d->name, i, io->state, ret);
1490 io->request_type = ASOK_REQ_NONE;
1491 some_unreachable = 1;
1496 done: for(i = 0; i < g_num_daemons; ++i)
1498 cconn_close(io_array + i);
1500 if(some_unreachable)
1502 DEBUG("ceph plugin: cconn_main_loop: some Ceph daemons were unreachable.");
1506 DEBUG("ceph plugin: cconn_main_loop: reached all Ceph daemons :)");
1511 static int ceph_read(void)
1513 return cconn_main_loop(ASOK_REQ_DATA);
1516 /******* lifecycle *******/
1517 static int ceph_init(void)
1520 DEBUG("ceph plugin: ceph_init");
1521 ceph_daemons_print();
1523 ret = cconn_main_loop(ASOK_REQ_VERSION);
1525 return (ret) ? ret : 0;
1528 static int ceph_shutdown(void)
1531 for(i = 0; i < g_num_daemons; ++i)
1533 ceph_daemon_free(g_daemons[i]);
1538 for(i = 0; i < last_idx; i++)
1540 sfree(last_poll_data[i]);
1542 sfree(last_poll_data);
1543 last_poll_data = NULL;
1545 DEBUG("ceph plugin: finished ceph_shutdown");
1549 void module_register(void)
1551 plugin_register_complex_config("ceph", ceph_config);
1552 plugin_register_init("ceph", ceph_init);
1553 plugin_register_read("ceph", ceph_read);
1554 plugin_register_shutdown("ceph", ceph_shutdown);