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
33 #include <arpa/inet.h>
36 #include <yajl/yajl_parse.h>
37 #if HAVE_YAJL_YAJL_VERSION_H
38 #include <yajl/yajl_version.h>
49 #include <sys/types.h>
55 #define RETRY_AVGCOUNT -1
57 #if defined(YAJL_MAJOR) && (YAJL_MAJOR > 1)
58 # define HAVE_YAJL_V2 1
61 #define RETRY_ON_EINTR(ret, expr) \
71 /** Timeout interval in seconds */
72 #define CEPH_TIMEOUT_INTERVAL 1
74 /** Maximum path length for a UNIX domain socket on this system */
75 #define UNIX_DOMAIN_SOCK_PATH_MAX (sizeof(((struct sockaddr_un*)0)->sun_path))
77 /** Yajl callback returns */
78 #define CEPH_CB_CONTINUE 1
79 #define CEPH_CB_ABORT 0
82 typedef size_t yajl_len_t;
84 typedef unsigned int yajl_len_t;
87 /** Number of types for ceph defined in types.db */
88 #define CEPH_DSET_TYPES_NUM 3
89 /** ceph types enum */
95 DSET_TYPE_UNFOUND = 1000
98 /** Valid types for ceph defined in types.db */
99 const char * ceph_dset_types [CEPH_DSET_TYPES_NUM] =
100 {"ceph_latency", "ceph_bytes", "ceph_rate"};
102 /******* ceph_daemon *******/
105 /** Version of the admin_socket interface */
108 char name[DATA_MAX_NAME_LEN];
110 /** Path to the socket that we use to talk to the ceph daemon */
111 char asok_path[UNIX_DOMAIN_SOCK_PATH_MAX];
113 /** Number of counters */
115 /** Track ds types */
117 /** Track ds names to match with types */
121 * Keep track of last data for latency values so we can calculate rate
124 struct last_data **last_poll_data;
125 /** index of last poll data */
129 /******* JSON parsing *******/
130 typedef int (*node_handler_t)(void *, const char*, const char*);
132 /** Track state and handler while parsing JSON */
135 node_handler_t handler;
138 char key[DATA_MAX_NAME_LEN];
140 } state[YAJL_MAX_DEPTH];
143 typedef struct yajl_struct yajl_struct;
145 enum perfcounter_type_d
147 PERFCOUNTER_LATENCY = 0x4, 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 int g_num_daemons = 0;
170 * A set of data that we build up in memory while parsing the JSON.
174 /** ceph daemon we are processing data for*/
175 struct ceph_daemon *d;
176 /** track avgcount across counters for avgcount/sum latency pairs */
178 /** current index of counters - used to get type of counter */
180 /** do we already have an avgcount for latency pair */
183 * similar to index, but current index of latency type counters -
184 * used to get last poll data of counter
188 * values list - maintain across counters since
189 * host/plugin/plugin instance are always the same
195 * A set of count/sum pairs to keep track of latency types and get difference
196 * between this poll data and last poll data.
200 char ds_name[DATA_MAX_NAME_LEN];
205 /******* network I/O *******/
208 CSTATE_UNCONNECTED = 0,
209 CSTATE_WRITE_REQUEST,
217 ASOK_REQ_VERSION = 0,
220 ASOK_REQ_NONE = 1000,
225 /** The Ceph daemon that we're talking to */
226 struct ceph_daemon *d;
229 uint32_t request_type;
231 /** The connection state */
234 /** The socket we use to talk to this daemon */
237 /** The amount of data remaining to read / write. */
240 /** Length of the JSON to read */
243 /** Buffer containing JSON data */
246 /** Keep data important to yajl processing */
247 struct yajl_struct yajl;
250 static int ceph_cb_null(void *ctx)
252 return CEPH_CB_CONTINUE;
255 static int ceph_cb_boolean(void *ctx, int bool_val)
257 return CEPH_CB_CONTINUE;
261 ceph_cb_number(void *ctx, const char *number_val, yajl_len_t number_len)
263 yajl_struct *yajl = (yajl_struct*)ctx;
264 char buffer[number_len+1];
265 int i, latency_type = 0, result;
268 memcpy(buffer, number_val, number_len);
269 buffer[sizeof(buffer) - 1] = 0;
271 ssnprintf(key, yajl->state[0].key_len, "%s", yajl->state[0].key);
272 for(i = 1; i < yajl->depth; i++)
274 if((i == yajl->depth-1) && ((strcmp(yajl->state[i].key,"avgcount") == 0)
275 || (strcmp(yajl->state[i].key,"sum") == 0)))
277 if(convert_special_metrics)
280 * Special case for filestore:JournalWrBytes. For some reason,
281 * Ceph schema encodes this as a count/sum pair while all
282 * other "Bytes" data (excluding used/capacity bytes for OSD
283 * space) uses a single "Derive" type. To spare further
284 * confusion, keep this KPI as the same type of other "Bytes".
285 * Instead of keeping an "average" or "rate", use the "sum" in
286 * the pair and assign that to the derive value.
288 if((strcmp(yajl->state[i-1].key, "journal_wr_bytes") == 0) &&
289 (strcmp(yajl->state[i-2].key,"filestore") == 0) &&
290 (strcmp(yajl->state[i].key,"avgcount") == 0))
292 DEBUG("ceph plugin: Skipping avgcount for filestore.JournalWrBytes");
293 yajl->depth = (yajl->depth - 1);
294 return CEPH_CB_CONTINUE;
297 //probably a avgcount/sum pair. if not - we'll try full key later
301 strncat(key, ".", 1);
302 strncat(key, yajl->state[i].key, yajl->state[i].key_len+1);
305 result = yajl->handler(yajl->handler_arg, buffer, key);
307 if((result == RETRY_AVGCOUNT) && latency_type)
309 strncat(key, ".", 1);
310 strncat(key, yajl->state[yajl->depth-1].key,
311 yajl->state[yajl->depth-1].key_len+1);
312 result = yajl->handler(yajl->handler_arg, buffer, key);
315 if(result == -ENOMEM)
317 ERROR("ceph plugin: memory allocation failed");
318 return CEPH_CB_ABORT;
321 yajl->depth = (yajl->depth - 1);
322 return CEPH_CB_CONTINUE;
325 static int ceph_cb_string(void *ctx, const unsigned char *string_val,
326 yajl_len_t string_len)
328 return CEPH_CB_CONTINUE;
331 static int ceph_cb_start_map(void *ctx)
333 return CEPH_CB_CONTINUE;
337 ceph_cb_map_key(void *ctx, const unsigned char *key, yajl_len_t string_len)
339 yajl_struct *yajl = (yajl_struct*)ctx;
341 if((yajl->depth+1) >= YAJL_MAX_DEPTH)
343 ERROR("ceph plugin: depth exceeds max, aborting.");
344 return CEPH_CB_ABORT;
347 char buffer[string_len+1];
349 memcpy(buffer, key, string_len);
350 buffer[sizeof(buffer) - 1] = 0;
352 snprintf(yajl->state[yajl->depth].key, sizeof(buffer), "%s", buffer);
353 yajl->state[yajl->depth].key_len = sizeof(buffer);
354 yajl->depth = (yajl->depth + 1);
356 return CEPH_CB_CONTINUE;
359 static int ceph_cb_end_map(void *ctx)
361 yajl_struct *yajl = (yajl_struct*)ctx;
363 yajl->depth = (yajl->depth - 1);
364 return CEPH_CB_CONTINUE;
367 static int ceph_cb_start_array(void *ctx)
369 return CEPH_CB_CONTINUE;
372 static int ceph_cb_end_array(void *ctx)
374 return CEPH_CB_CONTINUE;
377 static yajl_callbacks callbacks = {
391 static void ceph_daemon_print(const struct ceph_daemon *d)
393 DEBUG("ceph plugin: name=%s, asok_path=%s", d->name, d->asok_path);
396 static void ceph_daemons_print(void)
399 for(i = 0; i < g_num_daemons; ++i)
401 ceph_daemon_print(g_daemons[i]);
405 static void ceph_daemon_free(struct ceph_daemon *d)
408 for(; i < d->last_idx; i++)
410 sfree(d->last_poll_data[i]);
412 sfree(d->last_poll_data);
413 d->last_poll_data = NULL;
415 for(i = 0; i < d->ds_num; i++)
417 sfree(d->ds_names[i]);
424 /* compact_ds_name removed the special characters ":", "_", "-" and "+" from the
425 * intput string. Characters following these special characters are capitalized.
426 * Trailing "+" and "-" characters are replaces with the strings "Plus" and
428 static int compact_ds_name (char *buffer, size_t buffer_size, char const *src)
433 size_t ptr_size = buffer_size;
434 _Bool append_plus = 0;
435 _Bool append_minus = 0;
437 if ((buffer == NULL) || (buffer_size <= strlen ("Minus")) || (src == NULL))
440 src_copy = strdup (src);
441 src_len = strlen(src);
443 /* Remove trailing "+" and "-". */
444 if (src_copy[src_len - 1] == '+')
448 src_copy[src_len] = 0;
450 else if (src_copy[src_len - 1] == '-')
454 src_copy[src_len] = 0;
457 /* Split at special chars, capitalize first character, append to buffer. */
458 char *dummy = src_copy;
460 char *save_ptr = NULL;
461 while ((token = strtok_r (dummy, ":_-+", &save_ptr)) != NULL)
467 token[0] = toupper ((int) token[0]);
469 assert (ptr_size > 1);
471 len = strlen (token);
476 assert (len < ptr_size);
478 sstrncpy (ptr, token, len + 1);
487 /* Append "Plus" or "Minus" if "+" or "-" has been stripped above. */
488 if (append_plus || append_minus)
490 char const *append = "Plus";
494 size_t offset = buffer_size - (strlen (append) + 1);
495 if (offset > strlen (buffer))
496 offset = strlen (buffer);
498 sstrncpy (buffer + offset, append, buffer_size - offset);
505 static _Bool has_suffix (char const *str, char const *suffix)
507 size_t str_len = strlen (str);
508 size_t suffix_len = strlen (suffix);
511 if (suffix_len > str_len)
513 offset = str_len - suffix_len;
515 if (strcmp (str + offset, suffix) == 0)
521 /* count_parts returns the number of elements a "foo.bar.baz" style key has. */
522 static size_t count_parts (char const *key)
525 size_t parts_num = 0;
527 for (ptr = key; ptr != NULL; ptr = strchr (ptr + 1, '.'))
534 * Parse key to remove "type" if this is for schema and initiate compaction
536 static int parse_keys (char *buffer, size_t buffer_size, const char *key_str)
538 char tmp[2 * buffer_size];
540 if (buffer == NULL || buffer_size == 0 || key_str == NULL || strlen (key_str) == 0)
543 if ((count_parts (key_str) > 2) && has_suffix (key_str, ".type"))
545 /* strip ".type" suffix iff the key has more than two parts. */
546 size_t sz = strlen (key_str) - strlen (".type") + 1;
548 if (sz > sizeof (tmp))
550 sstrncpy (tmp, key_str, sz);
554 sstrncpy (tmp, key_str, sizeof (tmp));
557 return compact_ds_name (buffer, buffer_size, tmp);
561 * while parsing ceph admin socket schema, save counter name and type for later
564 static int ceph_daemon_add_ds_entry(struct ceph_daemon *d, const char *name,
568 char ds_name[DATA_MAX_NAME_LEN];
569 memset(ds_name, 0, sizeof(ds_name));
571 if(convert_special_metrics)
574 * Special case for filestore:JournalWrBytes. For some reason, Ceph
575 * schema encodes this as a count/sum pair while all other "Bytes" data
576 * (excluding used/capacity bytes for OSD space) uses a single "Derive"
577 * type. To spare further confusion, keep this KPI as the same type of
578 * other "Bytes". Instead of keeping an "average" or "rate", use the
579 * "sum" in the pair and assign that to the derive value.
581 if((strcmp(name,"filestore.journal_wr_bytes.type") == 0))
587 d->ds_names = realloc(d->ds_names, sizeof(char *) * (d->ds_num + 1));
593 d->ds_types = realloc(d->ds_types, sizeof(uint32_t) * (d->ds_num + 1));
599 d->ds_names[d->ds_num] = malloc(sizeof(char) * DATA_MAX_NAME_LEN);
600 if(!d->ds_names[d->ds_num])
605 type = (pc_type & PERFCOUNTER_DERIVE) ? DSET_RATE :
606 ((pc_type & PERFCOUNTER_LATENCY) ? DSET_LATENCY : DSET_BYTES);
607 d->ds_types[d->ds_num] = type;
609 if (parse_keys(ds_name, sizeof (ds_name), name))
614 sstrncpy(d->ds_names[d->ds_num], ds_name, DATA_MAX_NAME_LEN -1);
615 d->ds_num = (d->ds_num + 1);
620 /******* ceph_config *******/
621 static int cc_handle_str(struct oconfig_item_s *item, char *dest, int dest_len)
624 if(item->values_num != 1)
628 if(item->values[0].type != OCONFIG_TYPE_STRING)
632 val = item->values[0].value.string;
633 if(snprintf(dest, dest_len, "%s", val) > (dest_len - 1))
635 ERROR("ceph plugin: configuration parameter '%s' is too long.\n",
637 return -ENAMETOOLONG;
642 static int cc_handle_bool(struct oconfig_item_s *item, int *dest)
644 if(item->values_num != 1)
649 if(item->values[0].type != OCONFIG_TYPE_BOOLEAN)
654 *dest = (item->values[0].value.boolean) ? 1 : 0;
658 static int cc_add_daemon_config(oconfig_item_t *ci)
661 struct ceph_daemon *nd, cd;
662 struct ceph_daemon **tmp;
663 memset(&cd, 0, sizeof(struct ceph_daemon));
665 if((ci->values_num != 1) || (ci->values[0].type != OCONFIG_TYPE_STRING))
667 WARNING("ceph plugin: `Daemon' blocks need exactly one string "
672 ret = cc_handle_str(ci, cd.name, DATA_MAX_NAME_LEN);
678 for(i=0; i < ci->children_num; i++)
680 oconfig_item_t *child = ci->children + i;
682 if(strcasecmp("SocketPath", child->key) == 0)
684 ret = cc_handle_str(child, cd.asok_path, sizeof(cd.asok_path));
692 WARNING("ceph plugin: ignoring unknown option %s", child->key);
695 if(cd.name[0] == '\0')
697 ERROR("ceph plugin: you must configure a daemon name.\n");
700 else if(cd.asok_path[0] == '\0')
702 ERROR("ceph plugin(name=%s): you must configure an administrative "
703 "socket path.\n", cd.name);
706 else if(!((cd.asok_path[0] == '/') ||
707 (cd.asok_path[0] == '.' && cd.asok_path[1] == '/')))
709 ERROR("ceph plugin(name=%s): administrative socket paths must begin "
710 "with '/' or './' Can't parse: '%s'\n", cd.name, cd.asok_path);
714 tmp = realloc(g_daemons, (g_num_daemons+1) * sizeof(*g_daemons));
717 /* The positive return value here indicates that this is a
718 * runtime error, not a configuration error. */
723 nd = malloc(sizeof(*nd));
728 memcpy(nd, &cd, sizeof(*nd));
729 g_daemons[g_num_daemons++] = nd;
733 static int ceph_config(oconfig_item_t *ci)
737 for(i = 0; i < ci->children_num; ++i)
739 oconfig_item_t *child = ci->children + i;
740 if(strcasecmp("Daemon", child->key) == 0)
742 ret = cc_add_daemon_config(child);
745 ERROR("ceph plugin: Couldn't allocate memory");
750 //process other daemons and ignore this one
754 else if(strcasecmp("LongRunAvgLatency", child->key) == 0)
756 ret = cc_handle_bool(child, &long_run_latency_avg);
762 else if(strcasecmp("ConvertSpecialMetricTypes", child->key) == 0)
764 ret = cc_handle_bool(child, &convert_special_metrics);
772 WARNING("ceph plugin: ignoring unknown option %s", child->key);
779 * Parse JSON and get error message if present
782 traverse_json(const unsigned char *json, uint32_t json_len, yajl_handle hand)
784 yajl_status status = yajl_parse(hand, json, json_len);
789 case yajl_status_error:
790 msg = yajl_get_error(hand, /* verbose = */ 1,
791 /* jsonText = */ (unsigned char *) json,
792 (unsigned int) json_len);
793 ERROR ("ceph plugin: yajl_parse failed: %s", msg);
794 yajl_free_error(hand, msg);
796 case yajl_status_client_canceled:
804 * Add entry for each counter while parsing schema
807 node_handler_define_schema(void *arg, const char *val, const char *key)
809 struct ceph_daemon *d = (struct ceph_daemon *) arg;
812 return ceph_daemon_add_ds_entry(d, key, pc_type);
816 * Latency counter does not yet have an entry in last poll data - add it.
818 static int add_last(struct ceph_daemon *d, const char *ds_n, double cur_sum,
821 d->last_poll_data[d->last_idx] = malloc(1 * sizeof(struct last_data));
822 if(!d->last_poll_data[d->last_idx])
826 sstrncpy(d->last_poll_data[d->last_idx]->ds_name,ds_n,
827 sizeof(d->last_poll_data[d->last_idx]->ds_name));
828 d->last_poll_data[d->last_idx]->last_sum = cur_sum;
829 d->last_poll_data[d->last_idx]->last_count = cur_count;
830 d->last_idx = (d->last_idx + 1);
835 * Update latency counter or add new entry if it doesn't exist
837 static int update_last(struct ceph_daemon *d, const char *ds_n, int index,
838 double cur_sum, uint64_t cur_count)
840 if((d->last_idx > index) && (strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0))
842 d->last_poll_data[index]->last_sum = cur_sum;
843 d->last_poll_data[index]->last_count = cur_count;
847 if(!d->last_poll_data)
849 d->last_poll_data = malloc(1 * sizeof(struct last_data *));
850 if(!d->last_poll_data)
857 struct last_data **tmp_last = realloc(d->last_poll_data,
858 ((d->last_idx+1) * sizeof(struct last_data *)));
863 d->last_poll_data = tmp_last;
865 return add_last(d, ds_n, cur_sum, cur_count);
869 * If using index guess failed (shouldn't happen, but possible if counters
870 * get rearranged), resort to searching for counter name
872 static int backup_search_for_last_avg(struct ceph_daemon *d, const char *ds_n)
875 for(; i < d->last_idx; i++)
877 if(strcmp(d->last_poll_data[i]->ds_name, ds_n) == 0)
886 * Calculate average b/t current data and last poll data
887 * if last poll data exists
889 static double get_last_avg(struct ceph_daemon *d, const char *ds_n, int index,
890 double cur_sum, uint64_t cur_count)
892 double result = -1.1, sum_delt = 0.0;
893 uint64_t count_delt = 0;
895 if(d->last_idx > index)
897 if(strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0)
901 //test previous index
902 else if((index > 0) && (strcmp(d->last_poll_data[index-1]->ds_name, ds_n) == 0))
904 tmp_index = (index - 1);
908 tmp_index = backup_search_for_last_avg(d, ds_n);
911 if((tmp_index > -1) && (cur_count > d->last_poll_data[tmp_index]->last_count))
913 sum_delt = (cur_sum - d->last_poll_data[tmp_index]->last_sum);
914 count_delt = (cur_count - d->last_poll_data[tmp_index]->last_count);
915 result = (sum_delt / count_delt);
923 if(update_last(d, ds_n, tmp_index, cur_sum, cur_count) == -ENOMEM)
931 * If using index guess failed, resort to searching for counter name
933 static uint32_t backup_search_for_type(struct ceph_daemon *d, char *ds_name)
936 for(; idx < d->ds_num; idx++)
938 if(strcmp(d->ds_names[idx], ds_name) == 0)
940 return d->ds_types[idx];
943 return DSET_TYPE_UNFOUND;
947 * Process counter data and dispatch values
949 static int node_handler_fetch_data(void *arg, const char *val, const char *key)
954 struct values_tmp *vtmp = (struct values_tmp*) arg;
955 uint32_t type = DSET_TYPE_UNFOUND;
956 int index = vtmp->index;
958 char ds_name[DATA_MAX_NAME_LEN];
959 memset(ds_name, 0, sizeof(ds_name));
961 if (parse_keys (ds_name, sizeof (ds_name), key))
966 if(index >= vtmp->d->ds_num)
968 //don't overflow bounds of array
969 index = (vtmp->d->ds_num - 1);
973 * counters should remain in same order we parsed schema... we maintain the
974 * index variable to keep track of current point in list of counters. first
975 * use index to guess point in array for retrieving type. if that doesn't
976 * work, use the old way to get the counter type
978 if(strcmp(ds_name, vtmp->d->ds_names[index]) == 0)
981 type = vtmp->d->ds_types[index];
983 else if((index > 0) && (strcmp(ds_name, vtmp->d->ds_names[index-1]) == 0))
986 type = vtmp->d->ds_types[index-1];
989 if(type == DSET_TYPE_UNFOUND)
991 //couldn't find right type by guessing, check the old way
992 type = backup_search_for_type(vtmp->d, ds_name);
998 if(vtmp->avgcount_exists == -1)
1000 sscanf(val, "%" PRIu64, &vtmp->avgcount);
1001 vtmp->avgcount_exists = 0;
1002 //return after saving avgcount - don't dispatch value
1003 //until latency calculation
1009 sscanf(val, "%lf", &sum);
1011 if(vtmp->avgcount == 0)
1016 /** User wants latency values as long run avg */
1017 if(long_run_latency_avg)
1019 result = (sum / vtmp->avgcount);
1023 result = get_last_avg(vtmp->d, ds_name, vtmp->latency_index, sum, vtmp->avgcount);
1024 if(result == -ENOMEM)
1031 vtmp->avgcount_exists = -1;
1032 vtmp->latency_index = (vtmp->latency_index + 1);
1036 sscanf(val, "%lf", &tmp_d);
1040 sscanf(val, "%" PRIu64, &tmp_u);
1043 case DSET_TYPE_UNFOUND:
1045 ERROR("ceph plugin: ds %s was not properly initialized.", ds_name);
1049 sstrncpy(vtmp->vlist.type, ceph_dset_types[type], sizeof(vtmp->vlist.type));
1050 sstrncpy(vtmp->vlist.type_instance, ds_name, sizeof(vtmp->vlist.type_instance));
1051 vtmp->vlist.values = &uv;
1052 vtmp->vlist.values_len = 1;
1054 vtmp->index = (vtmp->index + 1);
1055 plugin_dispatch_values(&vtmp->vlist);
1060 static int cconn_connect(struct cconn *io)
1062 struct sockaddr_un address;
1064 if(io->state != CSTATE_UNCONNECTED)
1066 ERROR("ceph plugin: cconn_connect: io->state != CSTATE_UNCONNECTED");
1069 fd = socket(PF_UNIX, SOCK_STREAM, 0);
1073 ERROR("ceph plugin: cconn_connect: socket(PF_UNIX, SOCK_STREAM, 0) "
1074 "failed: error %d", err);
1077 memset(&address, 0, sizeof(struct sockaddr_un));
1078 address.sun_family = AF_UNIX;
1079 snprintf(address.sun_path, sizeof(address.sun_path), "%s",
1082 connect(fd, (struct sockaddr *) &address, sizeof(struct sockaddr_un)));
1085 ERROR("ceph plugin: cconn_connect: connect(%d) failed: error %d",
1091 flags = fcntl(fd, F_GETFL, 0);
1092 if(fcntl(fd, F_SETFL, flags | O_NONBLOCK) != 0)
1095 ERROR("ceph plugin: cconn_connect: fcntl(%d, O_NONBLOCK) error %d",
1101 io->state = CSTATE_WRITE_REQUEST;
1108 static void cconn_close(struct cconn *io)
1110 io->state = CSTATE_UNCONNECTED;
1114 RETRY_ON_EINTR(res, close(io->asok));
1123 /* Process incoming JSON counter data */
1125 cconn_process_data(struct cconn *io, yajl_struct *yajl, yajl_handle hand)
1128 struct values_tmp *vtmp = calloc(1, sizeof(struct values_tmp) * 1);
1134 vtmp->vlist = (value_list_t)VALUE_LIST_INIT;
1135 sstrncpy(vtmp->vlist.host, hostname_g, sizeof(vtmp->vlist.host));
1136 sstrncpy(vtmp->vlist.plugin, "ceph", sizeof(vtmp->vlist.plugin));
1137 sstrncpy(vtmp->vlist.plugin_instance, io->d->name, sizeof(vtmp->vlist.plugin_instance));
1140 vtmp->avgcount_exists = -1;
1141 vtmp->latency_index = 0;
1143 yajl->handler_arg = vtmp;
1144 ret = traverse_json(io->json, io->json_len, hand);
1150 * Initiate JSON parsing and print error if one occurs
1152 static int cconn_process_json(struct cconn *io)
1154 if((io->request_type != ASOK_REQ_DATA) &&
1155 (io->request_type != ASOK_REQ_SCHEMA))
1164 hand = yajl_alloc(&callbacks,
1166 /* alloc funcs = */ NULL,
1168 /* alloc funcs = */ NULL, NULL,
1170 /* context = */ (void *)(&io->yajl));
1174 ERROR ("ceph plugin: yajl_alloc failed.");
1180 switch(io->request_type)
1183 io->yajl.handler = node_handler_fetch_data;
1184 result = cconn_process_data(io, &io->yajl, hand);
1186 case ASOK_REQ_SCHEMA:
1187 //init daemon specific variables
1189 io->d->last_idx = 0;
1190 io->d->last_poll_data = NULL;
1191 io->yajl.handler = node_handler_define_schema;
1192 io->yajl.handler_arg = io->d;
1193 result = traverse_json(io->json, io->json_len, hand);
1203 status = yajl_complete_parse(hand);
1205 status = yajl_parse_complete(hand);
1208 if (status != yajl_status_ok)
1210 unsigned char *errmsg = yajl_get_error (hand, /* verbose = */ 0,
1211 /* jsonText = */ NULL, /* jsonTextLen = */ 0);
1212 ERROR ("ceph plugin: yajl_parse_complete failed: %s",
1214 yajl_free_error (hand, errmsg);
1224 static int cconn_validate_revents(struct cconn *io, int revents)
1226 if(revents & POLLERR)
1228 ERROR("ceph plugin: cconn_validate_revents(name=%s): got POLLERR",
1234 case CSTATE_WRITE_REQUEST:
1235 return (revents & POLLOUT) ? 0 : -EINVAL;
1236 case CSTATE_READ_VERSION:
1237 case CSTATE_READ_AMT:
1238 case CSTATE_READ_JSON:
1239 return (revents & POLLIN) ? 0 : -EINVAL;
1241 ERROR("ceph plugin: cconn_validate_revents(name=%s) got to "
1242 "illegal state on line %d", io->d->name, __LINE__);
1247 /** Handle a network event for a connection */
1248 static int cconn_handle_event(struct cconn *io)
1253 case CSTATE_UNCONNECTED:
1254 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1255 "state on line %d", io->d->name, __LINE__);
1258 case CSTATE_WRITE_REQUEST:
1261 snprintf(cmd, sizeof(cmd), "%s%d%s", "{ \"prefix\": \"",
1262 io->request_type, "\" }\n");
1263 size_t cmd_len = strlen(cmd);
1265 write(io->asok, ((char*)&cmd) + io->amt, cmd_len - io->amt));
1266 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,amt=%d,ret=%d)",
1267 io->d->name, io->state, io->amt, ret);
1273 if(io->amt >= cmd_len)
1276 switch (io->request_type)
1278 case ASOK_REQ_VERSION:
1279 io->state = CSTATE_READ_VERSION;
1282 io->state = CSTATE_READ_AMT;
1288 case CSTATE_READ_VERSION:
1291 read(io->asok, ((char*)(&io->d->version)) + io->amt,
1292 sizeof(io->d->version) - io->amt));
1293 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1294 io->d->name, io->state, ret);
1300 if(io->amt >= sizeof(io->d->version))
1302 io->d->version = ntohl(io->d->version);
1303 if(io->d->version != 1)
1305 ERROR("ceph plugin: cconn_handle_event(name=%s) not "
1306 "expecting version %d!", io->d->name, io->d->version);
1309 DEBUG("ceph plugin: cconn_handle_event(name=%s): identified as "
1310 "version %d", io->d->name, io->d->version);
1313 io->request_type = ASOK_REQ_SCHEMA;
1317 case CSTATE_READ_AMT:
1320 read(io->asok, ((char*)(&io->json_len)) + io->amt,
1321 sizeof(io->json_len) - io->amt));
1322 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1323 io->d->name, io->state, ret);
1329 if(io->amt >= sizeof(io->json_len))
1331 io->json_len = ntohl(io->json_len);
1333 io->state = CSTATE_READ_JSON;
1334 io->json = calloc(1, io->json_len + 1);
1337 ERROR("ceph plugin: error callocing io->json");
1343 case CSTATE_READ_JSON:
1346 read(io->asok, io->json + io->amt, io->json_len - io->amt));
1347 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1348 io->d->name, io->state, ret);
1354 if(io->amt >= io->json_len)
1356 ret = cconn_process_json(io);
1362 io->request_type = ASOK_REQ_NONE;
1367 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1368 "state on line %d", io->d->name, __LINE__);
1373 static int cconn_prepare(struct cconn *io, struct pollfd* fds)
1376 if(io->request_type == ASOK_REQ_NONE)
1378 /* The request has already been serviced. */
1381 else if((io->request_type == ASOK_REQ_DATA) && (io->d->ds_num == 0))
1383 /* If there are no counters to report on, don't bother
1390 case CSTATE_UNCONNECTED:
1391 ret = cconn_connect(io);
1401 fds->events = POLLOUT;
1403 case CSTATE_WRITE_REQUEST:
1405 fds->events = POLLOUT;
1407 case CSTATE_READ_VERSION:
1408 case CSTATE_READ_AMT:
1409 case CSTATE_READ_JSON:
1411 fds->events = POLLIN;
1414 ERROR("ceph plugin: cconn_prepare(name=%s) got to illegal state "
1415 "on line %d", io->d->name, __LINE__);
1420 /** Returns the difference between two struct timevals in milliseconds.
1421 * On overflow, we return max/min int.
1423 static int milli_diff(const struct timeval *t1, const struct timeval *t2)
1426 int sec_diff = t1->tv_sec - t2->tv_sec;
1427 int usec_diff = t1->tv_usec - t2->tv_usec;
1428 ret = usec_diff / 1000;
1429 ret += (sec_diff * 1000);
1430 return (ret > INT_MAX) ? INT_MAX : ((ret < INT_MIN) ? INT_MIN : (int)ret);
1433 /** This handles the actual network I/O to talk to the Ceph daemons.
1435 static int cconn_main_loop(uint32_t request_type)
1437 int i, ret, some_unreachable = 0;
1438 struct timeval end_tv;
1439 struct cconn io_array[g_num_daemons];
1441 DEBUG("ceph plugin: entering cconn_main_loop(request_type = %d)", request_type);
1443 /* create cconn array */
1444 memset(io_array, 0, sizeof(io_array));
1445 for(i = 0; i < g_num_daemons; ++i)
1447 io_array[i].d = g_daemons[i];
1448 io_array[i].request_type = request_type;
1449 io_array[i].state = CSTATE_UNCONNECTED;
1452 /** Calculate the time at which we should give up */
1453 gettimeofday(&end_tv, NULL);
1454 end_tv.tv_sec += CEPH_TIMEOUT_INTERVAL;
1460 struct cconn *polled_io_array[g_num_daemons];
1461 struct pollfd fds[g_num_daemons];
1462 memset(fds, 0, sizeof(fds));
1464 for(i = 0; i < g_num_daemons; ++i)
1466 struct cconn *io = io_array + i;
1467 ret = cconn_prepare(io, fds + nfds);
1470 WARNING("ceph plugin: cconn_prepare(name=%s,i=%d,st=%d)=%d",
1471 io->d->name, i, io->state, ret);
1473 io->request_type = ASOK_REQ_NONE;
1474 some_unreachable = 1;
1478 polled_io_array[nfds++] = io_array + i;
1487 gettimeofday(&tv, NULL);
1488 diff = milli_diff(&end_tv, &tv);
1493 WARNING("ceph plugin: cconn_main_loop: timed out.");
1496 RETRY_ON_EINTR(ret, poll(fds, nfds, diff));
1499 ERROR("ceph plugin: poll(2) error: %d", ret);
1502 for(i = 0; i < nfds; ++i)
1504 struct cconn *io = polled_io_array[i];
1505 int revents = fds[i].revents;
1510 else if(cconn_validate_revents(io, revents))
1512 WARNING("ceph plugin: cconn(name=%s,i=%d,st=%d): "
1513 "revents validation error: "
1514 "revents=0x%08x", io->d->name, i, io->state, revents);
1516 io->request_type = ASOK_REQ_NONE;
1517 some_unreachable = 1;
1521 int ret = cconn_handle_event(io);
1524 WARNING("ceph plugin: cconn_handle_event(name=%s,"
1525 "i=%d,st=%d): error %d", io->d->name, i, io->state, ret);
1527 io->request_type = ASOK_REQ_NONE;
1528 some_unreachable = 1;
1533 done: for(i = 0; i < g_num_daemons; ++i)
1535 cconn_close(io_array + i);
1537 if(some_unreachable)
1539 DEBUG("ceph plugin: cconn_main_loop: some Ceph daemons were unreachable.");
1543 DEBUG("ceph plugin: cconn_main_loop: reached all Ceph daemons :)");
1548 static int ceph_read(void)
1550 return cconn_main_loop(ASOK_REQ_DATA);
1553 /******* lifecycle *******/
1554 static int ceph_init(void)
1557 ceph_daemons_print();
1559 ret = cconn_main_loop(ASOK_REQ_VERSION);
1561 return (ret) ? ret : 0;
1564 static int ceph_shutdown(void)
1567 for(i = 0; i < g_num_daemons; ++i)
1569 ceph_daemon_free(g_daemons[i]);
1574 DEBUG("ceph plugin: finished ceph_shutdown");
1578 void module_register(void)
1580 plugin_register_complex_config("ceph", ceph_config);
1581 plugin_register_init("ceph", ceph_init);
1582 plugin_register_read("ceph", ceph_read);
1583 plugin_register_shutdown("ceph", ceph_shutdown);