2 * collectd - src/ceph.c
3 * Copyright (C) 2011 New Dream Network
4 * Copyright (C) 2015 Florian octo Forster
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the
8 * Free Software Foundation; only version 2 of the License is applicable.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License for more details.
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 * Colin McCabe <cmccabe at alumni.cmu.edu>
21 * Dennis Zou <yunzou at cisco.com>
22 * Dan Ryder <daryder at cisco.com>
23 * Florian octo Forster <octo at collectd.org>
26 #define _DEFAULT_SOURCE
34 #include <arpa/inet.h>
37 #include <yajl/yajl_parse.h>
38 #if HAVE_YAJL_YAJL_VERSION_H
39 #include <yajl/yajl_version.h>
41 #ifdef HAVE_SYS_CAPABILITY_H
42 #include <sys/capability.h>
55 #include <sys/types.h>
59 #define RETRY_AVGCOUNT -1
61 #if defined(YAJL_MAJOR) && (YAJL_MAJOR > 1)
62 #define HAVE_YAJL_V2 1
65 #define RETRY_ON_EINTR(ret, expr) \
75 /** Timeout interval in seconds */
76 #define CEPH_TIMEOUT_INTERVAL 1
78 /** Maximum path length for a UNIX domain socket on this system */
79 #define UNIX_DOMAIN_SOCK_PATH_MAX (sizeof(((struct sockaddr_un *)0)->sun_path))
81 /** Yajl callback returns */
82 #define CEPH_CB_CONTINUE 1
83 #define CEPH_CB_ABORT 0
86 typedef size_t yajl_len_t;
88 typedef unsigned int yajl_len_t;
91 /** Number of types for ceph defined in types.db */
92 #define CEPH_DSET_TYPES_NUM 3
93 /** ceph types enum */
94 enum ceph_dset_type_d {
98 DSET_TYPE_UNFOUND = 1000
101 /** Valid types for ceph defined in types.db */
102 static const char *const ceph_dset_types[CEPH_DSET_TYPES_NUM] = {
103 "ceph_latency", "ceph_bytes", "ceph_rate"};
105 /******* ceph_daemon *******/
107 /** Version of the admin_socket interface */
110 char name[DATA_MAX_NAME_LEN];
112 /** Path to the socket that we use to talk to the ceph daemon */
113 char asok_path[UNIX_DOMAIN_SOCK_PATH_MAX];
115 /** Number of counters */
117 /** Track ds types */
119 /** Track ds names to match with types */
123 * Keep track of last data for latency values so we can calculate rate
126 struct last_data **last_poll_data;
127 /** index of last poll data */
131 /******* JSON parsing *******/
132 typedef int (*node_handler_t)(void *, const char *, const char *);
134 /** Track state and handler while parsing JSON */
136 node_handler_t handler;
140 char *stack[YAJL_MAX_DEPTH];
143 typedef struct yajl_struct yajl_struct;
145 enum perfcounter_type_d {
146 PERFCOUNTER_LATENCY = 0x4,
147 PERFCOUNTER_DERIVE = 0x8,
150 /** Give user option to use default (long run = since daemon started) avg */
151 static int long_run_latency_avg = 0;
154 * Give user option to use default type for special cases -
155 * filestore.journal_wr_bytes is currently only metric here. Ceph reports the
156 * type as a sum/count pair and will calculate it the same as a latency value.
157 * All other "bytes" metrics (excluding the used/capacity bytes for the OSD)
158 * use the DERIVE type. Unless user specifies to use given type, convert this
159 * metric to use DERIVE.
161 static int convert_special_metrics = 1;
163 /** Array of daemons to monitor */
164 static struct ceph_daemon **g_daemons = NULL;
166 /** Number of elements in g_daemons */
167 static size_t g_num_daemons = 0;
170 * 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.
198 char ds_name[DATA_MAX_NAME_LEN];
203 /******* network I/O *******/
205 CSTATE_UNCONNECTED = 0,
206 CSTATE_WRITE_REQUEST,
212 enum request_type_t {
213 ASOK_REQ_VERSION = 0,
216 ASOK_REQ_NONE = 1000,
220 /** The Ceph daemon that we're talking to */
221 struct ceph_daemon *d;
224 uint32_t request_type;
226 /** The connection state */
229 /** The socket we use to talk to this daemon */
232 /** The amount of data remaining to read / write. */
235 /** Length of the JSON to read */
238 /** Buffer containing JSON data */
241 /** Keep data important to yajl processing */
242 struct yajl_struct yajl;
245 static int ceph_cb_null(void *ctx) { return CEPH_CB_CONTINUE; }
247 static int ceph_cb_boolean(void *ctx, int bool_val) { return CEPH_CB_CONTINUE; }
249 #define BUFFER_ADD(dest, src) \
251 size_t dest_size = sizeof(dest); \
252 size_t dest_len = strlen(dest); \
253 if (dest_size > dest_len) { \
254 sstrncpy((dest) + dest_len, (src), dest_size - dest_len); \
256 (dest)[dest_size - 1] = 0; \
259 static int ceph_cb_number(void *ctx, const char *number_val,
260 yajl_len_t number_len) {
261 yajl_struct *state = (yajl_struct *)ctx;
262 char buffer[number_len + 1];
263 char key[2 * DATA_MAX_NAME_LEN] = {0};
264 _Bool latency_type = 0;
267 memcpy(buffer, number_val, number_len);
268 buffer[sizeof(buffer) - 1] = '\0';
270 for (size_t i = 0; i < state->depth; i++) {
271 if (state->stack[i] == NULL)
274 if (strlen(key) != 0)
275 BUFFER_ADD(key, ".");
276 BUFFER_ADD(key, state->stack[i]);
279 /* Special case for latency metrics. */
280 if ((strcmp("avgcount", state->key) == 0) ||
281 (strcmp("sum", state->key) == 0)) {
284 /* depth >= 2 => (stack[-1] != NULL && stack[-2] != NULL) */
285 assert((state->depth < 2) || ((state->stack[state->depth - 1] != NULL) &&
286 (state->stack[state->depth - 2] != NULL)));
288 /* Super-special case for filestore.journal_wr_bytes.avgcount: For
289 * some reason, Ceph schema encodes this as a count/sum pair while all
290 * other "Bytes" data (excluding used/capacity bytes for OSD space) uses
291 * a single "Derive" type. To spare further confusion, keep this KPI as
292 * the same type of other "Bytes". Instead of keeping an "average" or
293 * "rate", use the "sum" in the pair and assign that to the derive
295 if (convert_special_metrics && (state->depth >= 2) &&
296 (strcmp("filestore", state->stack[state->depth - 2]) == 0) &&
297 (strcmp("journal_wr_bytes", state->stack[state->depth - 1]) == 0) &&
298 (strcmp("avgcount", state->key) == 0)) {
299 DEBUG("ceph plugin: Skipping avgcount for filestore.JournalWrBytes");
300 return CEPH_CB_CONTINUE;
302 } else /* not a latency type */
304 BUFFER_ADD(key, ".");
305 BUFFER_ADD(key, state->key);
308 status = state->handler(state->handler_arg, buffer, key);
309 if ((status == RETRY_AVGCOUNT) && latency_type) {
310 /* Add previously skipped part of the key, either "avgcount" or "sum",
312 BUFFER_ADD(key, ".");
313 BUFFER_ADD(key, state->key);
315 status = state->handler(state->handler_arg, buffer, key);
319 ERROR("ceph plugin: JSON handler failed with status %d.", status);
320 return CEPH_CB_ABORT;
323 return CEPH_CB_CONTINUE;
326 static int ceph_cb_string(void *ctx, const unsigned char *string_val,
327 yajl_len_t string_len) {
328 return CEPH_CB_CONTINUE;
331 static int ceph_cb_start_map(void *ctx) {
332 yajl_struct *state = (yajl_struct *)ctx;
334 /* Push key to the stack */
335 if (state->depth == YAJL_MAX_DEPTH)
336 return CEPH_CB_ABORT;
338 state->stack[state->depth] = state->key;
342 return CEPH_CB_CONTINUE;
345 static int ceph_cb_end_map(void *ctx) {
346 yajl_struct *state = (yajl_struct *)ctx;
348 /* Pop key from the stack */
349 if (state->depth == 0)
350 return CEPH_CB_ABORT;
354 state->key = state->stack[state->depth];
355 state->stack[state->depth] = NULL;
357 return CEPH_CB_CONTINUE;
360 static int ceph_cb_map_key(void *ctx, const unsigned char *key,
361 yajl_len_t string_len) {
362 yajl_struct *state = (yajl_struct *)ctx;
363 size_t sz = ((size_t)string_len) + 1;
366 state->key = malloc(sz);
367 if (state->key == NULL) {
368 ERROR("ceph plugin: malloc failed.");
369 return CEPH_CB_ABORT;
372 memmove(state->key, key, sz - 1);
373 state->key[sz - 1] = 0;
375 return CEPH_CB_CONTINUE;
378 static int ceph_cb_start_array(void *ctx) { return CEPH_CB_CONTINUE; }
380 static int ceph_cb_end_array(void *ctx) { return CEPH_CB_CONTINUE; }
382 static yajl_callbacks callbacks = {ceph_cb_null,
394 static void ceph_daemon_print(const struct ceph_daemon *d) {
395 DEBUG("ceph plugin: name=%s, asok_path=%s", d->name, d->asok_path);
398 static void ceph_daemons_print(void) {
399 for (size_t i = 0; i < g_num_daemons; ++i) {
400 ceph_daemon_print(g_daemons[i]);
404 static void ceph_daemon_free(struct ceph_daemon *d) {
405 for (int i = 0; i < d->last_idx; i++) {
406 sfree(d->last_poll_data[i]);
408 sfree(d->last_poll_data);
409 d->last_poll_data = NULL;
412 for (int i = 0; i < d->ds_num; i++) {
413 sfree(d->ds_names[i]);
420 /* compact_ds_name removed the special characters ":", "_", "-" and "+" from the
421 * input string. Characters following these special characters are capitalized.
422 * Trailing "+" and "-" characters are replaces with the strings "Plus" and
424 static int compact_ds_name(char *buffer, size_t buffer_size, char const *src) {
428 size_t ptr_size = buffer_size;
429 _Bool append_plus = 0;
430 _Bool append_minus = 0;
432 if ((buffer == NULL) || (buffer_size <= strlen("Minus")) || (src == NULL))
435 src_copy = strdup(src);
436 src_len = strlen(src);
438 /* Remove trailing "+" and "-". */
439 if (src_copy[src_len - 1] == '+') {
442 src_copy[src_len] = 0;
443 } else if (src_copy[src_len - 1] == '-') {
446 src_copy[src_len] = 0;
449 /* Split at special chars, capitalize first character, append to buffer. */
450 char *dummy = src_copy;
452 char *save_ptr = NULL;
453 while ((token = strtok_r(dummy, ":_-+", &save_ptr)) != NULL) {
458 token[0] = toupper((int)token[0]);
460 assert(ptr_size > 1);
467 assert(len < ptr_size);
469 sstrncpy(ptr, token, len + 1);
478 /* Append "Plus" or "Minus" if "+" or "-" has been stripped above. */
479 if (append_plus || append_minus) {
480 char const *append = "Plus";
484 size_t offset = buffer_size - (strlen(append) + 1);
485 if (offset > strlen(buffer))
486 offset = strlen(buffer);
488 sstrncpy(buffer + offset, append, buffer_size - offset);
495 static _Bool has_suffix(char const *str, char const *suffix) {
496 size_t str_len = strlen(str);
497 size_t suffix_len = strlen(suffix);
500 if (suffix_len > str_len)
502 offset = str_len - suffix_len;
504 if (strcmp(str + offset, suffix) == 0)
510 /* count_parts returns the number of elements a "foo.bar.baz" style key has. */
511 static size_t count_parts(char const *key) {
512 size_t parts_num = 0;
514 for (const char *ptr = key; ptr != NULL; ptr = strchr(ptr + 1, '.'))
521 * Parse key to remove "type" if this is for schema and initiate compaction
523 static int parse_keys(char *buffer, size_t buffer_size, const char *key_str) {
524 char tmp[2 * buffer_size];
526 if (buffer == NULL || buffer_size == 0 || key_str == NULL ||
527 strlen(key_str) == 0)
530 if ((count_parts(key_str) > 2) && has_suffix(key_str, ".type")) {
531 /* strip ".type" suffix iff the key has more than two parts. */
532 size_t sz = strlen(key_str) - strlen(".type") + 1;
534 if (sz > sizeof(tmp))
536 sstrncpy(tmp, key_str, sz);
538 sstrncpy(tmp, key_str, sizeof(tmp));
541 return compact_ds_name(buffer, buffer_size, tmp);
545 * while parsing ceph admin socket schema, save counter name and type for later
548 static int ceph_daemon_add_ds_entry(struct ceph_daemon *d, const char *name,
551 char ds_name[DATA_MAX_NAME_LEN];
553 if (convert_special_metrics) {
555 * Special case for filestore:JournalWrBytes. For some reason, Ceph
556 * schema encodes this as a count/sum pair while all other "Bytes" data
557 * (excluding used/capacity bytes for OSD space) uses a single "Derive"
558 * type. To spare further confusion, keep this KPI as the same type of
559 * other "Bytes". Instead of keeping an "average" or "rate", use the
560 * "sum" in the pair and assign that to the derive value.
562 if ((strcmp(name, "filestore.journal_wr_bytes.type") == 0)) {
567 d->ds_names = realloc(d->ds_names, sizeof(char *) * (d->ds_num + 1));
572 d->ds_types = realloc(d->ds_types, sizeof(uint32_t) * (d->ds_num + 1));
577 d->ds_names[d->ds_num] = malloc(DATA_MAX_NAME_LEN);
578 if (!d->ds_names[d->ds_num]) {
582 type = (pc_type & PERFCOUNTER_DERIVE)
584 : ((pc_type & PERFCOUNTER_LATENCY) ? DSET_LATENCY : DSET_BYTES);
585 d->ds_types[d->ds_num] = type;
587 if (parse_keys(ds_name, sizeof(ds_name), name)) {
591 sstrncpy(d->ds_names[d->ds_num], ds_name, DATA_MAX_NAME_LEN - 1);
592 d->ds_num = (d->ds_num + 1);
597 /******* ceph_config *******/
598 static int cc_handle_str(struct oconfig_item_s *item, char *dest,
601 if (item->values_num != 1) {
604 if (item->values[0].type != OCONFIG_TYPE_STRING) {
607 val = item->values[0].value.string;
608 if (snprintf(dest, dest_len, "%s", val) > (dest_len - 1)) {
609 ERROR("ceph plugin: configuration parameter '%s' is too long.\n",
611 return -ENAMETOOLONG;
616 static int cc_handle_bool(struct oconfig_item_s *item, int *dest) {
617 if (item->values_num != 1) {
621 if (item->values[0].type != OCONFIG_TYPE_BOOLEAN) {
625 *dest = (item->values[0].value.boolean) ? 1 : 0;
629 static int cc_add_daemon_config(oconfig_item_t *ci) {
631 struct ceph_daemon *nd, cd = {0};
632 struct ceph_daemon **tmp;
634 if ((ci->values_num != 1) || (ci->values[0].type != OCONFIG_TYPE_STRING)) {
635 WARNING("ceph plugin: `Daemon' blocks need exactly one string "
640 ret = cc_handle_str(ci, cd.name, DATA_MAX_NAME_LEN);
645 for (int i = 0; i < ci->children_num; i++) {
646 oconfig_item_t *child = ci->children + i;
648 if (strcasecmp("SocketPath", child->key) == 0) {
649 ret = cc_handle_str(child, cd.asok_path, sizeof(cd.asok_path));
654 WARNING("ceph plugin: ignoring unknown option %s", child->key);
657 if (cd.name[0] == '\0') {
658 ERROR("ceph plugin: you must configure a daemon name.\n");
660 } else if (cd.asok_path[0] == '\0') {
661 ERROR("ceph plugin(name=%s): you must configure an administrative "
665 } else if (!((cd.asok_path[0] == '/') ||
666 (cd.asok_path[0] == '.' && cd.asok_path[1] == '/'))) {
667 ERROR("ceph plugin(name=%s): administrative socket paths must begin "
668 "with '/' or './' Can't parse: '%s'\n",
669 cd.name, cd.asok_path);
673 tmp = realloc(g_daemons, (g_num_daemons + 1) * sizeof(*g_daemons));
675 /* The positive return value here indicates that this is a
676 * runtime error, not a configuration error. */
681 nd = malloc(sizeof(*nd));
685 memcpy(nd, &cd, sizeof(*nd));
686 g_daemons[g_num_daemons] = nd;
691 static int ceph_config(oconfig_item_t *ci) {
694 for (int i = 0; i < ci->children_num; ++i) {
695 oconfig_item_t *child = ci->children + i;
696 if (strcasecmp("Daemon", child->key) == 0) {
697 ret = cc_add_daemon_config(child);
699 ERROR("ceph plugin: Couldn't allocate memory");
702 // process other daemons and ignore this one
705 } else if (strcasecmp("LongRunAvgLatency", child->key) == 0) {
706 ret = cc_handle_bool(child, &long_run_latency_avg);
710 } else if (strcasecmp("ConvertSpecialMetricTypes", child->key) == 0) {
711 ret = cc_handle_bool(child, &convert_special_metrics);
716 WARNING("ceph plugin: ignoring unknown option %s", child->key);
723 * Parse JSON and get error message if present
725 static int traverse_json(const unsigned char *json, uint32_t json_len,
727 yajl_status status = yajl_parse(hand, json, json_len);
731 case yajl_status_error:
732 msg = yajl_get_error(hand, /* verbose = */ 1,
733 /* jsonText = */ (unsigned char *)json,
734 (unsigned int)json_len);
735 ERROR("ceph plugin: yajl_parse failed: %s", msg);
736 yajl_free_error(hand, msg);
738 case yajl_status_client_canceled:
746 * Add entry for each counter while parsing schema
748 static int node_handler_define_schema(void *arg, const char *val,
750 struct ceph_daemon *d = (struct ceph_daemon *)arg;
753 return ceph_daemon_add_ds_entry(d, key, pc_type);
757 * Latency counter does not yet have an entry in last poll data - add it.
759 static int add_last(struct ceph_daemon *d, const char *ds_n, double cur_sum,
760 uint64_t cur_count) {
761 d->last_poll_data[d->last_idx] =
762 malloc(sizeof(*d->last_poll_data[d->last_idx]));
763 if (!d->last_poll_data[d->last_idx]) {
766 sstrncpy(d->last_poll_data[d->last_idx]->ds_name, ds_n,
767 sizeof(d->last_poll_data[d->last_idx]->ds_name));
768 d->last_poll_data[d->last_idx]->last_sum = cur_sum;
769 d->last_poll_data[d->last_idx]->last_count = cur_count;
770 d->last_idx = (d->last_idx + 1);
775 * Update latency counter or add new entry if it doesn't exist
777 static int update_last(struct ceph_daemon *d, const char *ds_n, int index,
778 double cur_sum, uint64_t cur_count) {
779 if ((d->last_idx > index) &&
780 (strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0)) {
781 d->last_poll_data[index]->last_sum = cur_sum;
782 d->last_poll_data[index]->last_count = cur_count;
786 if (!d->last_poll_data) {
787 d->last_poll_data = malloc(sizeof(*d->last_poll_data));
788 if (!d->last_poll_data) {
792 struct last_data **tmp_last = realloc(
793 d->last_poll_data, ((d->last_idx + 1) * sizeof(struct last_data *)));
797 d->last_poll_data = tmp_last;
799 return add_last(d, ds_n, cur_sum, cur_count);
803 * If using index guess failed (shouldn't happen, but possible if counters
804 * get rearranged), resort to searching for counter name
806 static int backup_search_for_last_avg(struct ceph_daemon *d, const char *ds_n) {
807 for (int i = 0; i < d->last_idx; i++) {
808 if (strcmp(d->last_poll_data[i]->ds_name, ds_n) == 0) {
816 * Calculate average b/t current data and last poll data
817 * if last poll data exists
819 static double get_last_avg(struct ceph_daemon *d, const char *ds_n, int index,
820 double cur_sum, uint64_t cur_count) {
821 double result = -1.1, sum_delt = 0.0;
822 uint64_t count_delt = 0;
824 if (d->last_idx > index) {
825 if (strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0) {
828 // test previous index
829 else if ((index > 0) &&
830 (strcmp(d->last_poll_data[index - 1]->ds_name, ds_n) == 0)) {
831 tmp_index = (index - 1);
833 tmp_index = backup_search_for_last_avg(d, ds_n);
836 if ((tmp_index > -1) &&
837 (cur_count > d->last_poll_data[tmp_index]->last_count)) {
838 sum_delt = (cur_sum - d->last_poll_data[tmp_index]->last_sum);
839 count_delt = (cur_count - d->last_poll_data[tmp_index]->last_count);
840 result = (sum_delt / count_delt);
844 if (result == -1.1) {
847 if (update_last(d, ds_n, tmp_index, cur_sum, cur_count) == -ENOMEM) {
854 * If using index guess failed, resort to searching for counter name
856 static uint32_t backup_search_for_type(struct ceph_daemon *d, char *ds_name) {
857 for (int i = 0; i < d->ds_num; i++) {
858 if (strcmp(d->ds_names[i], ds_name) == 0) {
859 return d->ds_types[i];
862 return DSET_TYPE_UNFOUND;
866 * Process counter data and dispatch values
868 static int node_handler_fetch_data(void *arg, const char *val,
873 struct values_tmp *vtmp = (struct values_tmp *)arg;
874 uint32_t type = DSET_TYPE_UNFOUND;
875 int index = vtmp->index;
877 char ds_name[DATA_MAX_NAME_LEN];
879 if (parse_keys(ds_name, sizeof(ds_name), key)) {
883 if (index >= vtmp->d->ds_num) {
884 // don't overflow bounds of array
885 index = (vtmp->d->ds_num - 1);
889 * counters should remain in same order we parsed schema... we maintain the
890 * index variable to keep track of current point in list of counters. first
891 * use index to guess point in array for retrieving type. if that doesn't
892 * work, use the old way to get the counter type
894 if (strcmp(ds_name, vtmp->d->ds_names[index]) == 0) {
896 type = vtmp->d->ds_types[index];
897 } else if ((index > 0) &&
898 (strcmp(ds_name, vtmp->d->ds_names[index - 1]) == 0)) {
900 type = vtmp->d->ds_types[index - 1];
903 if (type == DSET_TYPE_UNFOUND) {
904 // couldn't find right type by guessing, check the old way
905 type = backup_search_for_type(vtmp->d, ds_name);
910 if (vtmp->avgcount_exists == -1) {
911 sscanf(val, "%" PRIu64, &vtmp->avgcount);
912 vtmp->avgcount_exists = 0;
913 // return after saving avgcount - don't dispatch value
914 // until latency calculation
918 sscanf(val, "%lf", &sum);
920 if (vtmp->avgcount == 0) {
924 /** User wants latency values as long run avg */
925 if (long_run_latency_avg) {
926 result = (sum / vtmp->avgcount);
928 result = get_last_avg(vtmp->d, ds_name, vtmp->latency_index, sum,
930 if (result == -ENOMEM) {
936 vtmp->avgcount_exists = -1;
937 vtmp->latency_index = (vtmp->latency_index + 1);
941 sscanf(val, "%lf", &tmp_d);
945 sscanf(val, "%" PRIu64, &tmp_u);
948 case DSET_TYPE_UNFOUND:
950 ERROR("ceph plugin: ds %s was not properly initialized.", ds_name);
954 sstrncpy(vtmp->vlist.type, ceph_dset_types[type], sizeof(vtmp->vlist.type));
955 sstrncpy(vtmp->vlist.type_instance, ds_name,
956 sizeof(vtmp->vlist.type_instance));
957 vtmp->vlist.values = &uv;
958 vtmp->vlist.values_len = 1;
960 vtmp->index = (vtmp->index + 1);
961 plugin_dispatch_values(&vtmp->vlist);
966 static int cconn_connect(struct cconn *io) {
967 struct sockaddr_un address = {0};
969 if (io->state != CSTATE_UNCONNECTED) {
970 ERROR("ceph plugin: cconn_connect: io->state != CSTATE_UNCONNECTED");
973 fd = socket(PF_UNIX, SOCK_STREAM, 0);
976 ERROR("ceph plugin: cconn_connect: socket(PF_UNIX, SOCK_STREAM, 0) "
981 address.sun_family = AF_UNIX;
982 snprintf(address.sun_path, sizeof(address.sun_path), "%s", io->d->asok_path);
983 RETRY_ON_EINTR(err, connect(fd, (struct sockaddr *)&address,
984 sizeof(struct sockaddr_un)));
986 ERROR("ceph plugin: cconn_connect: connect(%d) failed: error %d", fd, err);
991 flags = fcntl(fd, F_GETFL, 0);
992 if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) != 0) {
994 ERROR("ceph plugin: cconn_connect: fcntl(%d, O_NONBLOCK) error %d", fd,
1000 io->state = CSTATE_WRITE_REQUEST;
1007 static void cconn_close(struct cconn *io) {
1008 io->state = CSTATE_UNCONNECTED;
1009 if (io->asok != -1) {
1011 RETRY_ON_EINTR(res, close(io->asok));
1020 /* Process incoming JSON counter data */
1021 static int cconn_process_data(struct cconn *io, yajl_struct *yajl,
1024 struct values_tmp *vtmp = calloc(1, sizeof(struct values_tmp) * 1);
1029 vtmp->vlist = (value_list_t)VALUE_LIST_INIT;
1030 sstrncpy(vtmp->vlist.plugin, "ceph", sizeof(vtmp->vlist.plugin));
1031 sstrncpy(vtmp->vlist.plugin_instance, io->d->name,
1032 sizeof(vtmp->vlist.plugin_instance));
1035 vtmp->avgcount_exists = -1;
1036 vtmp->latency_index = 0;
1038 yajl->handler_arg = vtmp;
1039 ret = traverse_json(io->json, io->json_len, hand);
1045 * Initiate JSON parsing and print error if one occurs
1047 static int cconn_process_json(struct cconn *io) {
1048 if ((io->request_type != ASOK_REQ_DATA) &&
1049 (io->request_type != ASOK_REQ_SCHEMA)) {
1057 hand = yajl_alloc(&callbacks,
1059 /* alloc funcs = */ NULL,
1061 /* alloc funcs = */ NULL, NULL,
1063 /* context = */ (void *)(&io->yajl));
1066 ERROR("ceph plugin: yajl_alloc failed.");
1072 switch (io->request_type) {
1074 io->yajl.handler = node_handler_fetch_data;
1075 result = cconn_process_data(io, &io->yajl, hand);
1077 case ASOK_REQ_SCHEMA:
1078 // init daemon specific variables
1080 io->d->last_idx = 0;
1081 io->d->last_poll_data = NULL;
1082 io->yajl.handler = node_handler_define_schema;
1083 io->yajl.handler_arg = io->d;
1084 result = traverse_json(io->json, io->json_len, hand);
1093 status = yajl_complete_parse(hand);
1095 status = yajl_parse_complete(hand);
1098 if (status != yajl_status_ok) {
1099 unsigned char *errmsg =
1100 yajl_get_error(hand, /* verbose = */ 0,
1101 /* jsonText = */ NULL, /* jsonTextLen = */ 0);
1102 ERROR("ceph plugin: yajl_parse_complete failed: %s", (char *)errmsg);
1103 yajl_free_error(hand, errmsg);
1113 static int cconn_validate_revents(struct cconn *io, int revents) {
1114 if (revents & POLLERR) {
1115 ERROR("ceph plugin: cconn_validate_revents(name=%s): got POLLERR",
1119 switch (io->state) {
1120 case CSTATE_WRITE_REQUEST:
1121 return (revents & POLLOUT) ? 0 : -EINVAL;
1122 case CSTATE_READ_VERSION:
1123 case CSTATE_READ_AMT:
1124 case CSTATE_READ_JSON:
1125 return (revents & POLLIN) ? 0 : -EINVAL;
1127 ERROR("ceph plugin: cconn_validate_revents(name=%s) got to "
1128 "illegal state on line %d",
1129 io->d->name, __LINE__);
1134 /** Handle a network event for a connection */
1135 static int cconn_handle_event(struct cconn *io) {
1137 switch (io->state) {
1138 case CSTATE_UNCONNECTED:
1139 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1141 io->d->name, __LINE__);
1144 case CSTATE_WRITE_REQUEST: {
1146 snprintf(cmd, sizeof(cmd), "%s%d%s", "{ \"prefix\": \"", io->request_type,
1148 size_t cmd_len = strlen(cmd);
1150 ret, write(io->asok, ((char *)&cmd) + io->amt, cmd_len - io->amt));
1151 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,amt=%d,ret=%d)",
1152 io->d->name, io->state, io->amt, ret);
1157 if (io->amt >= cmd_len) {
1159 switch (io->request_type) {
1160 case ASOK_REQ_VERSION:
1161 io->state = CSTATE_READ_VERSION;
1164 io->state = CSTATE_READ_AMT;
1170 case CSTATE_READ_VERSION: {
1171 RETRY_ON_EINTR(ret, read(io->asok, ((char *)(&io->d->version)) + io->amt,
1172 sizeof(io->d->version) - io->amt));
1173 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1174 io->d->name, io->state, ret);
1179 if (io->amt >= sizeof(io->d->version)) {
1180 io->d->version = ntohl(io->d->version);
1181 if (io->d->version != 1) {
1182 ERROR("ceph plugin: cconn_handle_event(name=%s) not "
1183 "expecting version %d!",
1184 io->d->name, io->d->version);
1187 DEBUG("ceph plugin: cconn_handle_event(name=%s): identified as "
1189 io->d->name, io->d->version);
1192 io->request_type = ASOK_REQ_SCHEMA;
1196 case CSTATE_READ_AMT: {
1197 RETRY_ON_EINTR(ret, read(io->asok, ((char *)(&io->json_len)) + io->amt,
1198 sizeof(io->json_len) - io->amt));
1199 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1200 io->d->name, io->state, ret);
1205 if (io->amt >= sizeof(io->json_len)) {
1206 io->json_len = ntohl(io->json_len);
1208 io->state = CSTATE_READ_JSON;
1209 io->json = calloc(1, io->json_len + 1);
1211 ERROR("ceph plugin: error callocing io->json");
1217 case CSTATE_READ_JSON: {
1219 read(io->asok, io->json + io->amt, io->json_len - io->amt));
1220 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1221 io->d->name, io->state, ret);
1226 if (io->amt >= io->json_len) {
1227 ret = cconn_process_json(io);
1232 io->request_type = ASOK_REQ_NONE;
1237 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1239 io->d->name, __LINE__);
1244 static int cconn_prepare(struct cconn *io, struct pollfd *fds) {
1246 if (io->request_type == ASOK_REQ_NONE) {
1247 /* The request has already been serviced. */
1249 } else if ((io->request_type == ASOK_REQ_DATA) && (io->d->ds_num == 0)) {
1250 /* If there are no counters to report on, don't bother
1255 switch (io->state) {
1256 case CSTATE_UNCONNECTED:
1257 ret = cconn_connect(io);
1260 } else if (ret < 0) {
1264 fds->events = POLLOUT;
1266 case CSTATE_WRITE_REQUEST:
1268 fds->events = POLLOUT;
1270 case CSTATE_READ_VERSION:
1271 case CSTATE_READ_AMT:
1272 case CSTATE_READ_JSON:
1274 fds->events = POLLIN;
1277 ERROR("ceph plugin: cconn_prepare(name=%s) got to illegal state "
1279 io->d->name, __LINE__);
1284 /** Returns the difference between two struct timevals in milliseconds.
1285 * On overflow, we return max/min int.
1287 static int milli_diff(const struct timeval *t1, const struct timeval *t2) {
1289 int sec_diff = t1->tv_sec - t2->tv_sec;
1290 int usec_diff = t1->tv_usec - t2->tv_usec;
1291 ret = usec_diff / 1000;
1292 ret += (sec_diff * 1000);
1293 return (ret > INT_MAX) ? INT_MAX : ((ret < INT_MIN) ? INT_MIN : (int)ret);
1296 /** This handles the actual network I/O to talk to the Ceph daemons.
1298 static int cconn_main_loop(uint32_t request_type) {
1299 int ret, some_unreachable = 0;
1300 struct timeval end_tv;
1301 struct cconn io_array[g_num_daemons];
1303 DEBUG("ceph plugin: entering cconn_main_loop(request_type = %" PRIu32 ")",
1306 if (g_num_daemons < 1) {
1307 ERROR("ceph plugin: No daemons configured. See the \"Daemon\" config "
1312 /* create cconn array */
1313 for (size_t i = 0; i < g_num_daemons; i++) {
1314 io_array[i] = (struct cconn){
1316 .request_type = request_type,
1317 .state = CSTATE_UNCONNECTED,
1321 /** Calculate the time at which we should give up */
1322 gettimeofday(&end_tv, NULL);
1323 end_tv.tv_sec += CEPH_TIMEOUT_INTERVAL;
1328 struct cconn *polled_io_array[g_num_daemons];
1329 struct pollfd fds[g_num_daemons];
1330 memset(fds, 0, sizeof(fds));
1332 for (size_t i = 0; i < g_num_daemons; ++i) {
1333 struct cconn *io = io_array + i;
1334 ret = cconn_prepare(io, fds + nfds);
1336 WARNING("ceph plugin: cconn_prepare(name=%s,i=%zu,st=%d)=%d",
1337 io->d->name, i, io->state, ret);
1339 io->request_type = ASOK_REQ_NONE;
1340 some_unreachable = 1;
1341 } else if (ret == 1) {
1342 polled_io_array[nfds++] = io_array + i;
1350 gettimeofday(&tv, NULL);
1351 diff = milli_diff(&end_tv, &tv);
1355 WARNING("ceph plugin: cconn_main_loop: timed out.");
1358 RETRY_ON_EINTR(ret, poll(fds, nfds, diff));
1360 ERROR("ceph plugin: poll(2) error: %d", ret);
1363 for (int i = 0; i < nfds; ++i) {
1364 struct cconn *io = polled_io_array[i];
1365 int revents = fds[i].revents;
1369 } else if (cconn_validate_revents(io, revents)) {
1370 WARNING("ceph plugin: cconn(name=%s,i=%d,st=%d): "
1371 "revents validation error: "
1373 io->d->name, i, io->state, revents);
1375 io->request_type = ASOK_REQ_NONE;
1376 some_unreachable = 1;
1378 ret = cconn_handle_event(io);
1380 WARNING("ceph plugin: cconn_handle_event(name=%s,"
1381 "i=%d,st=%d): error %d",
1382 io->d->name, i, io->state, ret);
1384 io->request_type = ASOK_REQ_NONE;
1385 some_unreachable = 1;
1391 for (size_t i = 0; i < g_num_daemons; ++i) {
1392 cconn_close(io_array + i);
1394 if (some_unreachable) {
1395 DEBUG("ceph plugin: cconn_main_loop: some Ceph daemons were unreachable.");
1397 DEBUG("ceph plugin: cconn_main_loop: reached all Ceph daemons :)");
1402 static int ceph_read(void) { return cconn_main_loop(ASOK_REQ_DATA); }
1404 /******* lifecycle *******/
1405 static int ceph_init(void) {
1406 #if defined(HAVE_SYS_CAPABILITY_H) && defined(CAP_DAC_OVERRIDE)
1407 if (check_capability(CAP_DAC_OVERRIDE) != 0) {
1409 WARNING("ceph plugin: Running collectd as root, but the "
1410 "CAP_DAC_OVERRIDE capability is missing. The plugin's read "
1411 "function will probably fail. Is your init system dropping "
1415 "ceph plugin: collectd doesn't have the CAP_DAC_OVERRIDE "
1416 "capability. If you don't want to run collectd as root, try running "
1417 "\"setcap cap_dac_override=ep\" on the collectd binary.");
1421 ceph_daemons_print();
1423 if (g_num_daemons < 1) {
1424 ERROR("ceph plugin: No daemons configured. See the \"Daemon\" config "
1429 return cconn_main_loop(ASOK_REQ_VERSION);
1432 static int ceph_shutdown(void) {
1433 for (size_t i = 0; i < g_num_daemons; ++i) {
1434 ceph_daemon_free(g_daemons[i]);
1439 DEBUG("ceph plugin: finished ceph_shutdown");
1443 void module_register(void) {
1444 plugin_register_complex_config("ceph", ceph_config);
1445 plugin_register_init("ceph", ceph_init);
1446 plugin_register_read("ceph", ceph_read);
1447 plugin_register_shutdown("ceph", ceph_shutdown);