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>
24 #define _DEFAULT_SOURCE
31 #include <arpa/inet.h>
34 #include <yajl/yajl_parse.h>
35 #if HAVE_YAJL_YAJL_VERSION_H
36 #include <yajl/yajl_version.h>
46 #include <sys/socket.h>
48 #include <sys/types.h>
54 #define RETRY_AVGCOUNT -1
56 #if defined(YAJL_MAJOR) && (YAJL_MAJOR > 1)
57 # define HAVE_YAJL_V2 1
60 #define RETRY_ON_EINTR(ret, expr) \
70 /** Timeout interval in seconds */
71 #define CEPH_TIMEOUT_INTERVAL 1
73 /** Maximum path length for a UNIX domain socket on this system */
74 #define UNIX_DOMAIN_SOCK_PATH_MAX (sizeof(((struct sockaddr_un*)0)->sun_path))
76 /** Yajl callback returns */
77 #define CEPH_CB_CONTINUE 1
78 #define CEPH_CB_ABORT 0
81 typedef size_t yajl_len_t;
83 typedef unsigned int yajl_len_t;
86 /** Number of types for ceph defined in types.db */
87 #define CEPH_DSET_TYPES_NUM 3
88 /** ceph types enum */
94 DSET_TYPE_UNFOUND = 1000
97 /** Valid types for ceph defined in types.db */
98 const char * ceph_dset_types [CEPH_DSET_TYPES_NUM] =
99 {"ceph_latency", "ceph_bytes", "ceph_rate"};
101 /******* ceph_daemon *******/
104 /** Version of the admin_socket interface */
107 char name[DATA_MAX_NAME_LEN];
109 /** Path to the socket that we use to talk to the ceph daemon */
110 char asok_path[UNIX_DOMAIN_SOCK_PATH_MAX];
112 /** Number of counters */
114 /** Track ds types */
116 /** Track ds names to match with types */
120 * Keep track of last data for latency values so we can calculate rate
123 struct last_data **last_poll_data;
124 /** index of last poll data */
128 /******* JSON parsing *******/
129 typedef int (*node_handler_t)(void *, const char*, const char*);
131 /** Track state and handler while parsing JSON */
134 node_handler_t handler;
137 char key[DATA_MAX_NAME_LEN];
139 } state[YAJL_MAX_DEPTH];
142 typedef struct yajl_struct yajl_struct;
144 enum perfcounter_type_d
146 PERFCOUNTER_LATENCY = 0x4, PERFCOUNTER_DERIVE = 0x8,
149 /** Give user option to use default (long run = since daemon started) avg */
150 static int long_run_latency_avg = 0;
153 * Give user option to use default type for special cases -
154 * filestore.journal_wr_bytes is currently only metric here. Ceph reports the
155 * type as a sum/count pair and will calculate it the same as a latency value.
156 * All other "bytes" metrics (excluding the used/capacity bytes for the OSD)
157 * use the DERIVE type. Unless user specifies to use given type, convert this
158 * metric to use DERIVE.
160 static int convert_special_metrics = 1;
162 /** Array of daemons to monitor */
163 static struct ceph_daemon **g_daemons = NULL;
165 /** Number of elements in g_daemons */
166 static int g_num_daemons = 0;
169 * A set of data that we build up in memory while parsing the JSON.
173 /** ceph daemon we are processing data for*/
174 struct ceph_daemon *d;
175 /** track avgcount across counters for avgcount/sum latency pairs */
177 /** current index of counters - used to get type of counter */
179 /** do we already have an avgcount for latency pair */
182 * similar to index, but current index of latency type counters -
183 * used to get last poll data of counter
187 * values list - maintain across counters since
188 * host/plugin/plugin instance are always the same
194 * A set of count/sum pairs to keep track of latency types and get difference
195 * between this poll data and last poll data.
199 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->last_idx; i++)
409 sfree(d->last_poll_data[i]);
411 sfree(d->last_poll_data);
412 d->last_poll_data = NULL;
414 for(i = 0; i < d->ds_num; i++)
416 sfree(d->ds_names[i]);
424 * Compact ds name by removing special characters and trimming length to
425 * DATA_MAX_NAME_LEN if necessary
427 static void compact_ds_name(char *source, char *dest)
430 char *save_ptr = NULL, *tmp_ptr = source;
433 char tmp[DATA_MAX_NAME_LEN];
434 size_t key_chars_remaining = (DATA_MAX_NAME_LEN-1);
437 memset(tmp, 0, sizeof(tmp));
438 if(source == NULL || dest == NULL || source[0] == '\0' || dest[0] != '\0')
442 size_t src_len = strlen(source);
443 snprintf(len_str, sizeof(len_str), "%zu", src_len);
444 unsigned char append_status = 0x0;
445 append_status |= (source[src_len - 1] == '-') ? 0x1 : 0x0;
446 append_status |= (source[src_len - 1] == '+') ? 0x2 : 0x0;
447 while ((keys[keys_num] = strtok_r(tmp_ptr, ":_-+", &save_ptr)) != NULL)
450 /** capitalize 1st char **/
451 keys[keys_num][0] = toupper(keys[keys_num][0]);
458 /** concatenate each part of source string **/
459 for(i = 0; i < keys_num; i++)
461 strncat(tmp, keys[i], key_chars_remaining);
462 key_chars_remaining -= strlen(keys[i]);
464 tmp[DATA_MAX_NAME_LEN - 1] = '\0';
465 /** to coordinate limitation of length of type_instance
466 * we will truncate ds_name
467 * when the its length is more than
470 if(strlen(tmp) > DATA_MAX_NAME_LEN - 1)
472 append_status |= 0x4;
473 /** we should reserve space for
478 if(append_status & 0x1)
480 /** we should reserve space for
485 if(append_status & 0x2)
487 /** we should reserve space for
492 snprintf(dest, DATA_MAX_NAME_LEN - reserved, "%s", tmp);
493 offset = strlen(dest);
494 switch (append_status)
497 memcpy(dest + offset, "Minus", 5);
500 memcpy(dest + offset, "Plus", 5);
503 memcpy(dest + offset, len_str, 2);
506 memcpy(dest + offset, "Minus", 5);
507 memcpy(dest + offset + 5, len_str, 2);
510 memcpy(dest + offset, "Plus", 4);
511 memcpy(dest + offset + 4, len_str, 2);
519 * Parse key to remove "type" if this is for schema and initiate compaction
521 static int parse_keys(const char *key_str, char *ds_name)
524 size_t ds_name_len = 0;
526 * allow up to 100 characters before compaction - compact_ds_name will not
527 * allow more than DATA_MAX_NAME_LEN chars
529 int max_str_len = 100;
530 char tmp_ds_name[max_str_len];
531 memset(tmp_ds_name, 0, sizeof(tmp_ds_name));
532 if(ds_name == NULL || key_str == NULL || key_str[0] == '\0' ||
537 if((ptr = strchr(key_str, '.')) == NULL
538 || (rptr = strrchr(key_str, '.')) == NULL)
540 memcpy(tmp_ds_name, key_str, max_str_len - 1);
544 ds_name_len = (rptr - ptr) > max_str_len ? max_str_len : (rptr - ptr);
545 if((ds_name_len == 0) || strncmp(rptr + 1, "type", 4))
546 { /** copy whole key **/
547 memcpy(tmp_ds_name, key_str, max_str_len - 1);
550 {/** more than two keys **/
551 memcpy(tmp_ds_name, key_str, ((rptr - key_str) > (max_str_len - 1) ?
552 (max_str_len - 1) : (rptr - key_str)));
555 compact: compact_ds_name(tmp_ds_name, ds_name);
560 * while parsing ceph admin socket schema, save counter name and type for later
563 static int ceph_daemon_add_ds_entry(struct ceph_daemon *d, const char *name,
567 char ds_name[DATA_MAX_NAME_LEN];
568 memset(ds_name, 0, sizeof(ds_name));
570 if(convert_special_metrics)
573 * Special case for filestore:JournalWrBytes. For some reason, Ceph
574 * schema encodes this as a count/sum pair while all other "Bytes" data
575 * (excluding used/capacity bytes for OSD space) uses a single "Derive"
576 * type. To spare further confusion, keep this KPI as the same type of
577 * other "Bytes". Instead of keeping an "average" or "rate", use the
578 * "sum" in the pair and assign that to the derive value.
580 if((strcmp(name,"filestore.journal_wr_bytes.type") == 0))
586 d->ds_names = realloc(d->ds_names, sizeof(char *) * (d->ds_num + 1));
592 d->ds_types = realloc(d->ds_types, sizeof(uint32_t) * (d->ds_num + 1));
598 d->ds_names[d->ds_num] = malloc(sizeof(char) * DATA_MAX_NAME_LEN);
599 if(!d->ds_names[d->ds_num])
604 type = (pc_type & PERFCOUNTER_DERIVE) ? DSET_RATE :
605 ((pc_type & PERFCOUNTER_LATENCY) ? DSET_LATENCY : DSET_BYTES);
606 d->ds_types[d->ds_num] = type;
608 if(parse_keys(name, ds_name))
613 sstrncpy(d->ds_names[d->ds_num], ds_name, DATA_MAX_NAME_LEN -1);
614 d->ds_num = (d->ds_num + 1);
619 /******* ceph_config *******/
620 static int cc_handle_str(struct oconfig_item_s *item, char *dest, int dest_len)
623 if(item->values_num != 1)
627 if(item->values[0].type != OCONFIG_TYPE_STRING)
631 val = item->values[0].value.string;
632 if(snprintf(dest, dest_len, "%s", val) > (dest_len - 1))
634 ERROR("ceph plugin: configuration parameter '%s' is too long.\n",
636 return -ENAMETOOLONG;
641 static int cc_handle_bool(struct oconfig_item_s *item, int *dest)
643 if(item->values_num != 1)
648 if(item->values[0].type != OCONFIG_TYPE_BOOLEAN)
653 *dest = (item->values[0].value.boolean) ? 1 : 0;
657 static int cc_add_daemon_config(oconfig_item_t *ci)
660 struct ceph_daemon *nd, cd;
661 struct ceph_daemon **tmp;
662 memset(&cd, 0, sizeof(struct ceph_daemon));
664 if((ci->values_num != 1) || (ci->values[0].type != OCONFIG_TYPE_STRING))
666 WARNING("ceph plugin: `Daemon' blocks need exactly one string "
671 ret = cc_handle_str(ci, cd.name, DATA_MAX_NAME_LEN);
677 for(i=0; i < ci->children_num; i++)
679 oconfig_item_t *child = ci->children + i;
681 if(strcasecmp("SocketPath", child->key) == 0)
683 ret = cc_handle_str(child, cd.asok_path, sizeof(cd.asok_path));
691 WARNING("ceph plugin: ignoring unknown option %s", child->key);
694 if(cd.name[0] == '\0')
696 ERROR("ceph plugin: you must configure a daemon name.\n");
699 else if(cd.asok_path[0] == '\0')
701 ERROR("ceph plugin(name=%s): you must configure an administrative "
702 "socket path.\n", cd.name);
705 else if(!((cd.asok_path[0] == '/') ||
706 (cd.asok_path[0] == '.' && cd.asok_path[1] == '/')))
708 ERROR("ceph plugin(name=%s): administrative socket paths must begin "
709 "with '/' or './' Can't parse: '%s'\n", cd.name, cd.asok_path);
713 tmp = realloc(g_daemons, (g_num_daemons+1) * sizeof(*g_daemons));
716 /* The positive return value here indicates that this is a
717 * runtime error, not a configuration error. */
722 nd = malloc(sizeof(*nd));
727 memcpy(nd, &cd, sizeof(*nd));
728 g_daemons[g_num_daemons++] = nd;
732 static int ceph_config(oconfig_item_t *ci)
736 for(i = 0; i < ci->children_num; ++i)
738 oconfig_item_t *child = ci->children + i;
739 if(strcasecmp("Daemon", child->key) == 0)
741 ret = cc_add_daemon_config(child);
744 ERROR("ceph plugin: Couldn't allocate memory");
749 //process other daemons and ignore this one
753 else if(strcasecmp("LongRunAvgLatency", child->key) == 0)
755 ret = cc_handle_bool(child, &long_run_latency_avg);
761 else if(strcasecmp("ConvertSpecialMetricTypes", child->key) == 0)
763 ret = cc_handle_bool(child, &convert_special_metrics);
771 WARNING("ceph plugin: ignoring unknown option %s", child->key);
778 * Parse JSON and get error message if present
781 traverse_json(const unsigned char *json, uint32_t json_len, yajl_handle hand)
783 yajl_status status = yajl_parse(hand, json, json_len);
788 case yajl_status_error:
789 msg = yajl_get_error(hand, /* verbose = */ 1,
790 /* jsonText = */ (unsigned char *) json,
791 (unsigned int) json_len);
792 ERROR ("ceph plugin: yajl_parse failed: %s", msg);
793 yajl_free_error(hand, msg);
795 case yajl_status_client_canceled:
803 * Add entry for each counter while parsing schema
806 node_handler_define_schema(void *arg, const char *val, const char *key)
808 struct ceph_daemon *d = (struct ceph_daemon *) arg;
811 return ceph_daemon_add_ds_entry(d, key, pc_type);
815 * Latency counter does not yet have an entry in last poll data - add it.
817 static int add_last(struct ceph_daemon *d, const char *ds_n, double cur_sum,
820 d->last_poll_data[d->last_idx] = malloc(1 * sizeof(struct last_data));
821 if(!d->last_poll_data[d->last_idx])
825 sstrncpy(d->last_poll_data[d->last_idx]->ds_name,ds_n,
826 sizeof(d->last_poll_data[d->last_idx]->ds_name));
827 d->last_poll_data[d->last_idx]->last_sum = cur_sum;
828 d->last_poll_data[d->last_idx]->last_count = cur_count;
829 d->last_idx = (d->last_idx + 1);
834 * Update latency counter or add new entry if it doesn't exist
836 static int update_last(struct ceph_daemon *d, const char *ds_n, int index,
837 double cur_sum, uint64_t cur_count)
839 if((d->last_idx > index) && (strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0))
841 d->last_poll_data[index]->last_sum = cur_sum;
842 d->last_poll_data[index]->last_count = cur_count;
846 if(!d->last_poll_data)
848 d->last_poll_data = malloc(1 * sizeof(struct last_data *));
849 if(!d->last_poll_data)
856 struct last_data **tmp_last = realloc(d->last_poll_data,
857 ((d->last_idx+1) * sizeof(struct last_data *)));
862 d->last_poll_data = tmp_last;
864 return add_last(d, ds_n, cur_sum, cur_count);
868 * If using index guess failed (shouldn't happen, but possible if counters
869 * get rearranged), resort to searching for counter name
871 static int backup_search_for_last_avg(struct ceph_daemon *d, const char *ds_n)
874 for(; i < d->last_idx; i++)
876 if(strcmp(d->last_poll_data[i]->ds_name, ds_n) == 0)
885 * Calculate average b/t current data and last poll data
886 * if last poll data exists
888 static double get_last_avg(struct ceph_daemon *d, const char *ds_n, int index,
889 double cur_sum, uint64_t cur_count)
891 double result = -1.1, sum_delt = 0.0;
892 uint64_t count_delt = 0;
894 if(d->last_idx > index)
896 if(strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0)
900 //test previous index
901 else if((index > 0) && (strcmp(d->last_poll_data[index-1]->ds_name, ds_n) == 0))
903 tmp_index = (index - 1);
907 tmp_index = backup_search_for_last_avg(d, ds_n);
910 if((tmp_index > -1) && (cur_count > d->last_poll_data[tmp_index]->last_count))
912 sum_delt = (cur_sum - d->last_poll_data[tmp_index]->last_sum);
913 count_delt = (cur_count - d->last_poll_data[tmp_index]->last_count);
914 result = (sum_delt / count_delt);
922 if(update_last(d, ds_n, tmp_index, cur_sum, cur_count) == -ENOMEM)
930 * If using index guess failed, resort to searching for counter name
932 static uint32_t backup_search_for_type(struct ceph_daemon *d, char *ds_name)
935 for(; idx < d->ds_num; idx++)
937 if(strcmp(d->ds_names[idx], ds_name) == 0)
939 return d->ds_types[idx];
942 return DSET_TYPE_UNFOUND;
946 * Process counter data and dispatch values
948 static int node_handler_fetch_data(void *arg, const char *val, const char *key)
953 struct values_tmp *vtmp = (struct values_tmp*) arg;
954 uint32_t type = DSET_TYPE_UNFOUND;
955 int index = vtmp->index;
957 char ds_name[DATA_MAX_NAME_LEN];
958 memset(ds_name, 0, sizeof(ds_name));
960 if(parse_keys(key, ds_name))
965 if(index >= vtmp->d->ds_num)
967 //don't overflow bounds of array
968 index = (vtmp->d->ds_num - 1);
972 * counters should remain in same order we parsed schema... we maintain the
973 * index variable to keep track of current point in list of counters. first
974 * use index to guess point in array for retrieving type. if that doesn't
975 * work, use the old way to get the counter type
977 if(strcmp(ds_name, vtmp->d->ds_names[index]) == 0)
980 type = vtmp->d->ds_types[index];
982 else if((index > 0) && (strcmp(ds_name, vtmp->d->ds_names[index-1]) == 0))
985 type = vtmp->d->ds_types[index-1];
988 if(type == DSET_TYPE_UNFOUND)
990 //couldn't find right type by guessing, check the old way
991 type = backup_search_for_type(vtmp->d, ds_name);
997 if(vtmp->avgcount_exists == -1)
999 sscanf(val, "%" PRIu64, &vtmp->avgcount);
1000 vtmp->avgcount_exists = 0;
1001 //return after saving avgcount - don't dispatch value
1002 //until latency calculation
1008 sscanf(val, "%lf", &sum);
1010 if(vtmp->avgcount == 0)
1015 /** User wants latency values as long run avg */
1016 if(long_run_latency_avg)
1018 result = (sum / vtmp->avgcount);
1022 result = get_last_avg(vtmp->d, ds_name, vtmp->latency_index, sum, vtmp->avgcount);
1023 if(result == -ENOMEM)
1030 vtmp->avgcount_exists = -1;
1031 vtmp->latency_index = (vtmp->latency_index + 1);
1035 sscanf(val, "%lf", &tmp_d);
1039 sscanf(val, "%" PRIu64, &tmp_u);
1042 case DSET_TYPE_UNFOUND:
1044 ERROR("ceph plugin: ds %s was not properly initialized.", ds_name);
1048 sstrncpy(vtmp->vlist.type, ceph_dset_types[type], sizeof(vtmp->vlist.type));
1049 sstrncpy(vtmp->vlist.type_instance, ds_name, sizeof(vtmp->vlist.type_instance));
1050 vtmp->vlist.values = &uv;
1051 vtmp->vlist.values_len = 1;
1053 vtmp->index = (vtmp->index + 1);
1054 plugin_dispatch_values(&vtmp->vlist);
1059 static int cconn_connect(struct cconn *io)
1061 struct sockaddr_un address;
1063 if(io->state != CSTATE_UNCONNECTED)
1065 ERROR("ceph plugin: cconn_connect: io->state != CSTATE_UNCONNECTED");
1068 fd = socket(PF_UNIX, SOCK_STREAM, 0);
1072 ERROR("ceph plugin: cconn_connect: socket(PF_UNIX, SOCK_STREAM, 0) "
1073 "failed: error %d", err);
1076 memset(&address, 0, sizeof(struct sockaddr_un));
1077 address.sun_family = AF_UNIX;
1078 snprintf(address.sun_path, sizeof(address.sun_path), "%s",
1081 connect(fd, (struct sockaddr *) &address, sizeof(struct sockaddr_un)));
1084 ERROR("ceph plugin: cconn_connect: connect(%d) failed: error %d",
1089 flags = fcntl(fd, F_GETFL, 0);
1090 if(fcntl(fd, F_SETFL, flags | O_NONBLOCK) != 0)
1093 ERROR("ceph plugin: cconn_connect: fcntl(%d, O_NONBLOCK) error %d",
1098 io->state = CSTATE_WRITE_REQUEST;
1105 static void cconn_close(struct cconn *io)
1107 io->state = CSTATE_UNCONNECTED;
1111 RETRY_ON_EINTR(res, close(io->asok));
1120 /* Process incoming JSON counter data */
1122 cconn_process_data(struct cconn *io, yajl_struct *yajl, yajl_handle hand)
1125 struct values_tmp *vtmp = calloc(1, sizeof(struct values_tmp) * 1);
1131 vtmp->vlist = (value_list_t)VALUE_LIST_INIT;
1132 sstrncpy(vtmp->vlist.host, hostname_g, sizeof(vtmp->vlist.host));
1133 sstrncpy(vtmp->vlist.plugin, "ceph", sizeof(vtmp->vlist.plugin));
1134 sstrncpy(vtmp->vlist.plugin_instance, io->d->name, sizeof(vtmp->vlist.plugin_instance));
1137 vtmp->avgcount_exists = -1;
1138 vtmp->latency_index = 0;
1140 yajl->handler_arg = vtmp;
1141 ret = traverse_json(io->json, io->json_len, hand);
1147 * Initiate JSON parsing and print error if one occurs
1149 static int cconn_process_json(struct cconn *io)
1151 if((io->request_type != ASOK_REQ_DATA) &&
1152 (io->request_type != ASOK_REQ_SCHEMA))
1161 hand = yajl_alloc(&callbacks,
1163 /* alloc funcs = */ NULL,
1165 /* alloc funcs = */ NULL, NULL,
1167 /* context = */ (void *)(&io->yajl));
1171 ERROR ("ceph plugin: yajl_alloc failed.");
1177 switch(io->request_type)
1180 io->yajl.handler = node_handler_fetch_data;
1181 result = cconn_process_data(io, &io->yajl, hand);
1183 case ASOK_REQ_SCHEMA:
1184 //init daemon specific variables
1186 io->d->last_idx = 0;
1187 io->d->last_poll_data = NULL;
1188 io->yajl.handler = node_handler_define_schema;
1189 io->yajl.handler_arg = io->d;
1190 result = traverse_json(io->json, io->json_len, hand);
1200 status = yajl_complete_parse(hand);
1202 status = yajl_parse_complete(hand);
1205 if (status != yajl_status_ok)
1207 unsigned char *errmsg = yajl_get_error (hand, /* verbose = */ 0,
1208 /* jsonText = */ NULL, /* jsonTextLen = */ 0);
1209 ERROR ("ceph plugin: yajl_parse_complete failed: %s",
1211 yajl_free_error (hand, errmsg);
1221 static int cconn_validate_revents(struct cconn *io, int revents)
1223 if(revents & POLLERR)
1225 ERROR("ceph plugin: cconn_validate_revents(name=%s): got POLLERR",
1231 case CSTATE_WRITE_REQUEST:
1232 return (revents & POLLOUT) ? 0 : -EINVAL;
1233 case CSTATE_READ_VERSION:
1234 case CSTATE_READ_AMT:
1235 case CSTATE_READ_JSON:
1236 return (revents & POLLIN) ? 0 : -EINVAL;
1238 ERROR("ceph plugin: cconn_validate_revents(name=%s) got to "
1239 "illegal state on line %d", io->d->name, __LINE__);
1244 /** Handle a network event for a connection */
1245 static int cconn_handle_event(struct cconn *io)
1250 case CSTATE_UNCONNECTED:
1251 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1252 "state on line %d", io->d->name, __LINE__);
1255 case CSTATE_WRITE_REQUEST:
1258 snprintf(cmd, sizeof(cmd), "%s%d%s", "{ \"prefix\": \"",
1259 io->request_type, "\" }\n");
1260 size_t cmd_len = strlen(cmd);
1262 write(io->asok, ((char*)&cmd) + io->amt, cmd_len - io->amt));
1263 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,amt=%d,ret=%d)",
1264 io->d->name, io->state, io->amt, ret);
1270 if(io->amt >= cmd_len)
1273 switch (io->request_type)
1275 case ASOK_REQ_VERSION:
1276 io->state = CSTATE_READ_VERSION;
1279 io->state = CSTATE_READ_AMT;
1285 case CSTATE_READ_VERSION:
1288 read(io->asok, ((char*)(&io->d->version)) + io->amt,
1289 sizeof(io->d->version) - io->amt));
1290 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1291 io->d->name, io->state, ret);
1297 if(io->amt >= sizeof(io->d->version))
1299 io->d->version = ntohl(io->d->version);
1300 if(io->d->version != 1)
1302 ERROR("ceph plugin: cconn_handle_event(name=%s) not "
1303 "expecting version %d!", io->d->name, io->d->version);
1306 DEBUG("ceph plugin: cconn_handle_event(name=%s): identified as "
1307 "version %d", io->d->name, io->d->version);
1310 io->request_type = ASOK_REQ_SCHEMA;
1314 case CSTATE_READ_AMT:
1317 read(io->asok, ((char*)(&io->json_len)) + io->amt,
1318 sizeof(io->json_len) - io->amt));
1319 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1320 io->d->name, io->state, ret);
1326 if(io->amt >= sizeof(io->json_len))
1328 io->json_len = ntohl(io->json_len);
1330 io->state = CSTATE_READ_JSON;
1331 io->json = calloc(1, io->json_len + 1);
1334 ERROR("ceph plugin: error callocing io->json");
1340 case CSTATE_READ_JSON:
1343 read(io->asok, io->json + io->amt, io->json_len - io->amt));
1344 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1345 io->d->name, io->state, ret);
1351 if(io->amt >= io->json_len)
1353 ret = cconn_process_json(io);
1359 io->request_type = ASOK_REQ_NONE;
1364 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1365 "state on line %d", io->d->name, __LINE__);
1370 static int cconn_prepare(struct cconn *io, struct pollfd* fds)
1373 if(io->request_type == ASOK_REQ_NONE)
1375 /* The request has already been serviced. */
1378 else if((io->request_type == ASOK_REQ_DATA) && (io->d->ds_num == 0))
1380 /* If there are no counters to report on, don't bother
1387 case CSTATE_UNCONNECTED:
1388 ret = cconn_connect(io);
1398 fds->events = POLLOUT;
1400 case CSTATE_WRITE_REQUEST:
1402 fds->events = POLLOUT;
1404 case CSTATE_READ_VERSION:
1405 case CSTATE_READ_AMT:
1406 case CSTATE_READ_JSON:
1408 fds->events = POLLIN;
1411 ERROR("ceph plugin: cconn_prepare(name=%s) got to illegal state "
1412 "on line %d", io->d->name, __LINE__);
1417 /** Returns the difference between two struct timevals in milliseconds.
1418 * On overflow, we return max/min int.
1420 static int milli_diff(const struct timeval *t1, const struct timeval *t2)
1423 int sec_diff = t1->tv_sec - t2->tv_sec;
1424 int usec_diff = t1->tv_usec - t2->tv_usec;
1425 ret = usec_diff / 1000;
1426 ret += (sec_diff * 1000);
1427 return (ret > INT_MAX) ? INT_MAX : ((ret < INT_MIN) ? INT_MIN : (int)ret);
1430 /** This handles the actual network I/O to talk to the Ceph daemons.
1432 static int cconn_main_loop(uint32_t request_type)
1434 int i, ret, some_unreachable = 0;
1435 struct timeval end_tv;
1436 struct cconn io_array[g_num_daemons];
1438 DEBUG("ceph plugin: entering cconn_main_loop(request_type = %d)", request_type);
1440 /* create cconn array */
1441 memset(io_array, 0, sizeof(io_array));
1442 for(i = 0; i < g_num_daemons; ++i)
1444 io_array[i].d = g_daemons[i];
1445 io_array[i].request_type = request_type;
1446 io_array[i].state = CSTATE_UNCONNECTED;
1449 /** Calculate the time at which we should give up */
1450 gettimeofday(&end_tv, NULL);
1451 end_tv.tv_sec += CEPH_TIMEOUT_INTERVAL;
1457 struct cconn *polled_io_array[g_num_daemons];
1458 struct pollfd fds[g_num_daemons];
1459 memset(fds, 0, sizeof(fds));
1461 for(i = 0; i < g_num_daemons; ++i)
1463 struct cconn *io = io_array + i;
1464 ret = cconn_prepare(io, fds + nfds);
1467 WARNING("ceph plugin: cconn_prepare(name=%s,i=%d,st=%d)=%d",
1468 io->d->name, i, io->state, ret);
1470 io->request_type = ASOK_REQ_NONE;
1471 some_unreachable = 1;
1475 polled_io_array[nfds++] = io_array + i;
1484 gettimeofday(&tv, NULL);
1485 diff = milli_diff(&end_tv, &tv);
1490 WARNING("ceph plugin: cconn_main_loop: timed out.");
1493 RETRY_ON_EINTR(ret, poll(fds, nfds, diff));
1496 ERROR("ceph plugin: poll(2) error: %d", ret);
1499 for(i = 0; i < nfds; ++i)
1501 struct cconn *io = polled_io_array[i];
1502 int revents = fds[i].revents;
1507 else if(cconn_validate_revents(io, revents))
1509 WARNING("ceph plugin: cconn(name=%s,i=%d,st=%d): "
1510 "revents validation error: "
1511 "revents=0x%08x", io->d->name, i, io->state, revents);
1513 io->request_type = ASOK_REQ_NONE;
1514 some_unreachable = 1;
1518 int ret = cconn_handle_event(io);
1521 WARNING("ceph plugin: cconn_handle_event(name=%s,"
1522 "i=%d,st=%d): error %d", io->d->name, i, io->state, ret);
1524 io->request_type = ASOK_REQ_NONE;
1525 some_unreachable = 1;
1530 done: for(i = 0; i < g_num_daemons; ++i)
1532 cconn_close(io_array + i);
1534 if(some_unreachable)
1536 DEBUG("ceph plugin: cconn_main_loop: some Ceph daemons were unreachable.");
1540 DEBUG("ceph plugin: cconn_main_loop: reached all Ceph daemons :)");
1545 static int ceph_read(void)
1547 return cconn_main_loop(ASOK_REQ_DATA);
1550 /******* lifecycle *******/
1551 static int ceph_init(void)
1554 ceph_daemons_print();
1556 ret = cconn_main_loop(ASOK_REQ_VERSION);
1558 return (ret) ? ret : 0;
1561 static int ceph_shutdown(void)
1564 for(i = 0; i < g_num_daemons; ++i)
1566 ceph_daemon_free(g_daemons[i]);
1571 DEBUG("ceph plugin: finished ceph_shutdown");
1575 void module_register(void)
1577 plugin_register_complex_config("ceph", ceph_config);
1578 plugin_register_init("ceph", ceph_init);
1579 plugin_register_read("ceph", ceph_read);
1580 plugin_register_shutdown("ceph", ceph_shutdown);