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>
48 #include <sys/socket.h>
50 #include <sys/types.h>
56 #define RETRY_AVGCOUNT -1
58 #if defined(YAJL_MAJOR) && (YAJL_MAJOR > 1)
59 # define HAVE_YAJL_V2 1
62 #define RETRY_ON_EINTR(ret, expr) \
72 /** Timeout interval in seconds */
73 #define CEPH_TIMEOUT_INTERVAL 1
75 /** Maximum path length for a UNIX domain socket on this system */
76 #define UNIX_DOMAIN_SOCK_PATH_MAX (sizeof(((struct sockaddr_un*)0)->sun_path))
78 /** Yajl callback returns */
79 #define CEPH_CB_CONTINUE 1
80 #define CEPH_CB_ABORT 0
83 typedef size_t yajl_len_t;
85 typedef unsigned int yajl_len_t;
88 /** Number of types for ceph defined in types.db */
89 #define CEPH_DSET_TYPES_NUM 3
90 /** ceph types enum */
96 DSET_TYPE_UNFOUND = 1000
99 /** Valid types for ceph defined in types.db */
100 const char * ceph_dset_types [CEPH_DSET_TYPES_NUM] =
101 {"ceph_latency", "ceph_bytes", "ceph_rate"};
103 /******* ceph_daemon *******/
106 /** Version of the admin_socket interface */
109 char name[DATA_MAX_NAME_LEN];
111 /** Path to the socket that we use to talk to the ceph daemon */
112 char asok_path[UNIX_DOMAIN_SOCK_PATH_MAX];
114 /** Number of counters */
116 /** Track ds types */
118 /** Track ds names to match with types */
122 * Keep track of last data for latency values so we can calculate rate
125 struct last_data **last_poll_data;
126 /** index of last poll data */
130 /******* JSON parsing *******/
131 typedef int (*node_handler_t)(void *, const char*, const char*);
133 /** Track state and handler while parsing JSON */
136 node_handler_t handler;
139 char key[DATA_MAX_NAME_LEN];
141 } state[YAJL_MAX_DEPTH];
144 typedef struct yajl_struct yajl_struct;
146 enum perfcounter_type_d
148 PERFCOUNTER_LATENCY = 0x4, PERFCOUNTER_DERIVE = 0x8,
151 /** Give user option to use default (long run = since daemon started) avg */
152 static int long_run_latency_avg = 0;
155 * Give user option to use default type for special cases -
156 * filestore.journal_wr_bytes is currently only metric here. Ceph reports the
157 * type as a sum/count pair and will calculate it the same as a latency value.
158 * All other "bytes" metrics (excluding the used/capacity bytes for the OSD)
159 * use the DERIVE type. Unless user specifies to use given type, convert this
160 * metric to use DERIVE.
162 static int convert_special_metrics = 1;
164 /** Array of daemons to monitor */
165 static struct ceph_daemon **g_daemons = NULL;
167 /** Number of elements in g_daemons */
168 static int g_num_daemons = 0;
171 * A set of data that we build up in memory while parsing the JSON.
175 /** ceph daemon we are processing data for*/
176 struct ceph_daemon *d;
177 /** track avgcount across counters for avgcount/sum latency pairs */
179 /** current index of counters - used to get type of counter */
181 /** do we already have an avgcount for latency pair */
184 * similar to index, but current index of latency type counters -
185 * used to get last poll data of counter
189 * values list - maintain across counters since
190 * host/plugin/plugin instance are always the same
196 * A set of count/sum pairs to keep track of latency types and get difference
197 * between this poll data and last poll data.
201 char ds_name[DATA_MAX_NAME_LEN];
206 /******* network I/O *******/
209 CSTATE_UNCONNECTED = 0,
210 CSTATE_WRITE_REQUEST,
218 ASOK_REQ_VERSION = 0,
221 ASOK_REQ_NONE = 1000,
226 /** The Ceph daemon that we're talking to */
227 struct ceph_daemon *d;
230 uint32_t request_type;
232 /** The connection state */
235 /** The socket we use to talk to this daemon */
238 /** The amount of data remaining to read / write. */
241 /** Length of the JSON to read */
244 /** Buffer containing JSON data */
247 /** Keep data important to yajl processing */
248 struct yajl_struct yajl;
251 static int ceph_cb_null(void *ctx)
253 return CEPH_CB_CONTINUE;
256 static int ceph_cb_boolean(void *ctx, int bool_val)
258 return CEPH_CB_CONTINUE;
261 #define BUFFER_ADD(dest, src) do { \
262 size_t dest_size = sizeof (dest); \
263 strncat ((dest), (src), dest_size - strlen (dest)); \
264 (dest)[dest_size - 1] = 0; \
268 ceph_cb_number(void *ctx, const char *number_val, yajl_len_t number_len)
270 yajl_struct *yajl = (yajl_struct*)ctx;
271 char buffer[number_len+1];
273 char key[2 * DATA_MAX_NAME_LEN];
274 _Bool latency_type = 0;
276 memcpy(buffer, number_val, number_len);
277 buffer[sizeof(buffer) - 1] = 0;
279 sstrncpy (key, yajl->state[0].key, sizeof (key));
280 for (i = 1; i < yajl->depth; i++)
282 /* Special case for latency metrics. */
283 if ((i == yajl->depth-1)
284 && ((strcmp(yajl->state[i].key,"avgcount") == 0)
285 || (strcmp(yajl->state[i].key,"sum") == 0)))
287 /* Super-special case for filestore:JournalWrBytes. For some
288 * reason, Ceph schema encodes this as a count/sum pair while all
289 * other "Bytes" data (excluding used/capacity bytes for OSD space)
290 * uses a single "Derive" type. To spare further confusion, keep
291 * this KPI as the same type of other "Bytes". Instead of keeping
292 * an "average" or "rate", use the "sum" in the pair and assign
293 * that to the derive value. */
294 if (convert_special_metrics && (i >= 2)
295 && (strcmp("filestore", yajl->state[i-2].key) == 0)
296 && (strcmp("journal_wr_bytes", yajl->state[i-1].key) == 0)
297 && (strcmp("avgcount", yajl->state[i].key) == 0))
299 DEBUG("ceph plugin: Skipping avgcount for filestore.JournalWrBytes");
301 return CEPH_CB_CONTINUE;
304 /* Don't add "avgcount" or "sum" to the key just yet. If the
305 * handler function signals RETRY_AVGCOUNT, we'll add it and try
311 BUFFER_ADD (key, ".");
312 BUFFER_ADD (key, yajl->state[i].key);
315 status = yajl->handler(yajl->handler_arg, buffer, key);
316 if((status == RETRY_AVGCOUNT) && latency_type)
318 /* Add previously skipped part of the key, either "avgcount" or "sum",
320 BUFFER_ADD (key, ".");
321 BUFFER_ADD (key, yajl->state[yajl->depth-1].key);
323 status = yajl->handler(yajl->handler_arg, buffer, key);
326 if(status == -ENOMEM)
328 ERROR("ceph plugin: memory allocation failed");
329 return CEPH_CB_ABORT;
333 return CEPH_CB_CONTINUE;
336 static int ceph_cb_string(void *ctx, const unsigned char *string_val,
337 yajl_len_t string_len)
339 return CEPH_CB_CONTINUE;
342 static int ceph_cb_start_map(void *ctx)
344 return CEPH_CB_CONTINUE;
348 ceph_cb_map_key(void *ctx, const unsigned char *key, yajl_len_t string_len)
350 yajl_struct *yajl = (yajl_struct*)ctx;
352 if((yajl->depth+1) >= YAJL_MAX_DEPTH)
354 ERROR("ceph plugin: depth exceeds max, aborting.");
355 return CEPH_CB_ABORT;
358 char buffer[string_len+1];
360 memcpy(buffer, key, string_len);
361 buffer[sizeof(buffer) - 1] = 0;
363 snprintf(yajl->state[yajl->depth].key, sizeof(buffer), "%s", buffer);
364 yajl->state[yajl->depth].key_len = sizeof(buffer);
365 yajl->depth = (yajl->depth + 1);
367 return CEPH_CB_CONTINUE;
370 static int ceph_cb_end_map(void *ctx)
372 yajl_struct *yajl = (yajl_struct*)ctx;
374 yajl->depth = (yajl->depth - 1);
375 return CEPH_CB_CONTINUE;
378 static int ceph_cb_start_array(void *ctx)
380 return CEPH_CB_CONTINUE;
383 static int ceph_cb_end_array(void *ctx)
385 return CEPH_CB_CONTINUE;
388 static yajl_callbacks callbacks = {
402 static void ceph_daemon_print(const struct ceph_daemon *d)
404 DEBUG("ceph plugin: name=%s, asok_path=%s", d->name, d->asok_path);
407 static void ceph_daemons_print(void)
410 for(i = 0; i < g_num_daemons; ++i)
412 ceph_daemon_print(g_daemons[i]);
416 static void ceph_daemon_free(struct ceph_daemon *d)
419 for(; i < d->last_idx; i++)
421 sfree(d->last_poll_data[i]);
423 sfree(d->last_poll_data);
424 d->last_poll_data = NULL;
426 for(i = 0; i < d->ds_num; i++)
428 sfree(d->ds_names[i]);
435 /* compact_ds_name removed the special characters ":", "_", "-" and "+" from the
436 * intput string. Characters following these special characters are capitalized.
437 * Trailing "+" and "-" characters are replaces with the strings "Plus" and
439 static int compact_ds_name (char *buffer, size_t buffer_size, char const *src)
444 size_t ptr_size = buffer_size;
445 _Bool append_plus = 0;
446 _Bool append_minus = 0;
448 if ((buffer == NULL) || (buffer_size <= strlen ("Minus")) || (src == NULL))
451 src_copy = strdup (src);
452 src_len = strlen(src);
454 /* Remove trailing "+" and "-". */
455 if (src_copy[src_len - 1] == '+')
459 src_copy[src_len] = 0;
461 else if (src_copy[src_len - 1] == '-')
465 src_copy[src_len] = 0;
468 /* Split at special chars, capitalize first character, append to buffer. */
469 char *dummy = src_copy;
471 char *save_ptr = NULL;
472 while ((token = strtok_r (dummy, ":_-+", &save_ptr)) != NULL)
478 token[0] = toupper ((int) token[0]);
480 assert (ptr_size > 1);
482 len = strlen (token);
487 assert (len < ptr_size);
489 sstrncpy (ptr, token, len + 1);
498 /* Append "Plus" or "Minus" if "+" or "-" has been stripped above. */
499 if (append_plus || append_minus)
501 char const *append = "Plus";
505 size_t offset = buffer_size - (strlen (append) + 1);
506 if (offset > strlen (buffer))
507 offset = strlen (buffer);
509 sstrncpy (buffer + offset, append, buffer_size - offset);
516 static _Bool has_suffix (char const *str, char const *suffix)
518 size_t str_len = strlen (str);
519 size_t suffix_len = strlen (suffix);
522 if (suffix_len > str_len)
524 offset = str_len - suffix_len;
526 if (strcmp (str + offset, suffix) == 0)
532 /* count_parts returns the number of elements a "foo.bar.baz" style key has. */
533 static size_t count_parts (char const *key)
536 size_t parts_num = 0;
538 for (ptr = key; ptr != NULL; ptr = strchr (ptr + 1, '.'))
545 * Parse key to remove "type" if this is for schema and initiate compaction
547 static int parse_keys (char *buffer, size_t buffer_size, const char *key_str)
549 char tmp[2 * buffer_size];
551 if (buffer == NULL || buffer_size == 0 || key_str == NULL || strlen (key_str) == 0)
554 if ((count_parts (key_str) > 2) && has_suffix (key_str, ".type"))
556 /* strip ".type" suffix iff the key has more than two parts. */
557 size_t sz = strlen (key_str) - strlen (".type") + 1;
559 if (sz > sizeof (tmp))
561 sstrncpy (tmp, key_str, sz);
565 sstrncpy (tmp, key_str, sizeof (tmp));
568 return compact_ds_name (buffer, buffer_size, tmp);
572 * while parsing ceph admin socket schema, save counter name and type for later
575 static int ceph_daemon_add_ds_entry(struct ceph_daemon *d, const char *name,
579 char ds_name[DATA_MAX_NAME_LEN];
580 memset(ds_name, 0, sizeof(ds_name));
582 if(convert_special_metrics)
585 * Special case for filestore:JournalWrBytes. For some reason, Ceph
586 * schema encodes this as a count/sum pair while all other "Bytes" data
587 * (excluding used/capacity bytes for OSD space) uses a single "Derive"
588 * type. To spare further confusion, keep this KPI as the same type of
589 * other "Bytes". Instead of keeping an "average" or "rate", use the
590 * "sum" in the pair and assign that to the derive value.
592 if((strcmp(name,"filestore.journal_wr_bytes.type") == 0))
598 d->ds_names = realloc(d->ds_names, sizeof(char *) * (d->ds_num + 1));
604 d->ds_types = realloc(d->ds_types, sizeof(uint32_t) * (d->ds_num + 1));
610 d->ds_names[d->ds_num] = malloc(sizeof(char) * DATA_MAX_NAME_LEN);
611 if(!d->ds_names[d->ds_num])
616 type = (pc_type & PERFCOUNTER_DERIVE) ? DSET_RATE :
617 ((pc_type & PERFCOUNTER_LATENCY) ? DSET_LATENCY : DSET_BYTES);
618 d->ds_types[d->ds_num] = type;
620 if (parse_keys(ds_name, sizeof (ds_name), name))
625 sstrncpy(d->ds_names[d->ds_num], ds_name, DATA_MAX_NAME_LEN -1);
626 d->ds_num = (d->ds_num + 1);
631 /******* ceph_config *******/
632 static int cc_handle_str(struct oconfig_item_s *item, char *dest, int dest_len)
635 if(item->values_num != 1)
639 if(item->values[0].type != OCONFIG_TYPE_STRING)
643 val = item->values[0].value.string;
644 if(snprintf(dest, dest_len, "%s", val) > (dest_len - 1))
646 ERROR("ceph plugin: configuration parameter '%s' is too long.\n",
648 return -ENAMETOOLONG;
653 static int cc_handle_bool(struct oconfig_item_s *item, int *dest)
655 if(item->values_num != 1)
660 if(item->values[0].type != OCONFIG_TYPE_BOOLEAN)
665 *dest = (item->values[0].value.boolean) ? 1 : 0;
669 static int cc_add_daemon_config(oconfig_item_t *ci)
672 struct ceph_daemon *nd, cd;
673 struct ceph_daemon **tmp;
674 memset(&cd, 0, sizeof(struct ceph_daemon));
676 if((ci->values_num != 1) || (ci->values[0].type != OCONFIG_TYPE_STRING))
678 WARNING("ceph plugin: `Daemon' blocks need exactly one string "
683 ret = cc_handle_str(ci, cd.name, DATA_MAX_NAME_LEN);
689 for(i=0; i < ci->children_num; i++)
691 oconfig_item_t *child = ci->children + i;
693 if(strcasecmp("SocketPath", child->key) == 0)
695 ret = cc_handle_str(child, cd.asok_path, sizeof(cd.asok_path));
703 WARNING("ceph plugin: ignoring unknown option %s", child->key);
706 if(cd.name[0] == '\0')
708 ERROR("ceph plugin: you must configure a daemon name.\n");
711 else if(cd.asok_path[0] == '\0')
713 ERROR("ceph plugin(name=%s): you must configure an administrative "
714 "socket path.\n", cd.name);
717 else if(!((cd.asok_path[0] == '/') ||
718 (cd.asok_path[0] == '.' && cd.asok_path[1] == '/')))
720 ERROR("ceph plugin(name=%s): administrative socket paths must begin "
721 "with '/' or './' Can't parse: '%s'\n", cd.name, cd.asok_path);
725 tmp = realloc(g_daemons, (g_num_daemons+1) * sizeof(*g_daemons));
728 /* The positive return value here indicates that this is a
729 * runtime error, not a configuration error. */
734 nd = malloc(sizeof(*nd));
739 memcpy(nd, &cd, sizeof(*nd));
740 g_daemons[g_num_daemons++] = nd;
744 static int ceph_config(oconfig_item_t *ci)
748 for(i = 0; i < ci->children_num; ++i)
750 oconfig_item_t *child = ci->children + i;
751 if(strcasecmp("Daemon", child->key) == 0)
753 ret = cc_add_daemon_config(child);
756 ERROR("ceph plugin: Couldn't allocate memory");
761 //process other daemons and ignore this one
765 else if(strcasecmp("LongRunAvgLatency", child->key) == 0)
767 ret = cc_handle_bool(child, &long_run_latency_avg);
773 else if(strcasecmp("ConvertSpecialMetricTypes", child->key) == 0)
775 ret = cc_handle_bool(child, &convert_special_metrics);
783 WARNING("ceph plugin: ignoring unknown option %s", child->key);
790 * Parse JSON and get error message if present
793 traverse_json(const unsigned char *json, uint32_t json_len, yajl_handle hand)
795 yajl_status status = yajl_parse(hand, json, json_len);
800 case yajl_status_error:
801 msg = yajl_get_error(hand, /* verbose = */ 1,
802 /* jsonText = */ (unsigned char *) json,
803 (unsigned int) json_len);
804 ERROR ("ceph plugin: yajl_parse failed: %s", msg);
805 yajl_free_error(hand, msg);
807 case yajl_status_client_canceled:
815 * Add entry for each counter while parsing schema
818 node_handler_define_schema(void *arg, const char *val, const char *key)
820 struct ceph_daemon *d = (struct ceph_daemon *) arg;
823 return ceph_daemon_add_ds_entry(d, key, pc_type);
827 * Latency counter does not yet have an entry in last poll data - add it.
829 static int add_last(struct ceph_daemon *d, const char *ds_n, double cur_sum,
832 d->last_poll_data[d->last_idx] = malloc(1 * sizeof(struct last_data));
833 if(!d->last_poll_data[d->last_idx])
837 sstrncpy(d->last_poll_data[d->last_idx]->ds_name,ds_n,
838 sizeof(d->last_poll_data[d->last_idx]->ds_name));
839 d->last_poll_data[d->last_idx]->last_sum = cur_sum;
840 d->last_poll_data[d->last_idx]->last_count = cur_count;
841 d->last_idx = (d->last_idx + 1);
846 * Update latency counter or add new entry if it doesn't exist
848 static int update_last(struct ceph_daemon *d, const char *ds_n, int index,
849 double cur_sum, uint64_t cur_count)
851 if((d->last_idx > index) && (strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0))
853 d->last_poll_data[index]->last_sum = cur_sum;
854 d->last_poll_data[index]->last_count = cur_count;
858 if(!d->last_poll_data)
860 d->last_poll_data = malloc(1 * sizeof(struct last_data *));
861 if(!d->last_poll_data)
868 struct last_data **tmp_last = realloc(d->last_poll_data,
869 ((d->last_idx+1) * sizeof(struct last_data *)));
874 d->last_poll_data = tmp_last;
876 return add_last(d, ds_n, cur_sum, cur_count);
880 * If using index guess failed (shouldn't happen, but possible if counters
881 * get rearranged), resort to searching for counter name
883 static int backup_search_for_last_avg(struct ceph_daemon *d, const char *ds_n)
886 for(; i < d->last_idx; i++)
888 if(strcmp(d->last_poll_data[i]->ds_name, ds_n) == 0)
897 * Calculate average b/t current data and last poll data
898 * if last poll data exists
900 static double get_last_avg(struct ceph_daemon *d, const char *ds_n, int index,
901 double cur_sum, uint64_t cur_count)
903 double result = -1.1, sum_delt = 0.0;
904 uint64_t count_delt = 0;
906 if(d->last_idx > index)
908 if(strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0)
912 //test previous index
913 else if((index > 0) && (strcmp(d->last_poll_data[index-1]->ds_name, ds_n) == 0))
915 tmp_index = (index - 1);
919 tmp_index = backup_search_for_last_avg(d, ds_n);
922 if((tmp_index > -1) && (cur_count > d->last_poll_data[tmp_index]->last_count))
924 sum_delt = (cur_sum - d->last_poll_data[tmp_index]->last_sum);
925 count_delt = (cur_count - d->last_poll_data[tmp_index]->last_count);
926 result = (sum_delt / count_delt);
934 if(update_last(d, ds_n, tmp_index, cur_sum, cur_count) == -ENOMEM)
942 * If using index guess failed, resort to searching for counter name
944 static uint32_t backup_search_for_type(struct ceph_daemon *d, char *ds_name)
947 for(; idx < d->ds_num; idx++)
949 if(strcmp(d->ds_names[idx], ds_name) == 0)
951 return d->ds_types[idx];
954 return DSET_TYPE_UNFOUND;
958 * Process counter data and dispatch values
960 static int node_handler_fetch_data(void *arg, const char *val, const char *key)
965 struct values_tmp *vtmp = (struct values_tmp*) arg;
966 uint32_t type = DSET_TYPE_UNFOUND;
967 int index = vtmp->index;
969 char ds_name[DATA_MAX_NAME_LEN];
970 memset(ds_name, 0, sizeof(ds_name));
972 if (parse_keys (ds_name, sizeof (ds_name), key))
977 if(index >= vtmp->d->ds_num)
979 //don't overflow bounds of array
980 index = (vtmp->d->ds_num - 1);
984 * counters should remain in same order we parsed schema... we maintain the
985 * index variable to keep track of current point in list of counters. first
986 * use index to guess point in array for retrieving type. if that doesn't
987 * work, use the old way to get the counter type
989 if(strcmp(ds_name, vtmp->d->ds_names[index]) == 0)
992 type = vtmp->d->ds_types[index];
994 else if((index > 0) && (strcmp(ds_name, vtmp->d->ds_names[index-1]) == 0))
997 type = vtmp->d->ds_types[index-1];
1000 if(type == DSET_TYPE_UNFOUND)
1002 //couldn't find right type by guessing, check the old way
1003 type = backup_search_for_type(vtmp->d, ds_name);
1009 if(vtmp->avgcount_exists == -1)
1011 sscanf(val, "%" PRIu64, &vtmp->avgcount);
1012 vtmp->avgcount_exists = 0;
1013 //return after saving avgcount - don't dispatch value
1014 //until latency calculation
1020 sscanf(val, "%lf", &sum);
1022 if(vtmp->avgcount == 0)
1027 /** User wants latency values as long run avg */
1028 if(long_run_latency_avg)
1030 result = (sum / vtmp->avgcount);
1034 result = get_last_avg(vtmp->d, ds_name, vtmp->latency_index, sum, vtmp->avgcount);
1035 if(result == -ENOMEM)
1042 vtmp->avgcount_exists = -1;
1043 vtmp->latency_index = (vtmp->latency_index + 1);
1047 sscanf(val, "%lf", &tmp_d);
1051 sscanf(val, "%" PRIu64, &tmp_u);
1054 case DSET_TYPE_UNFOUND:
1056 ERROR("ceph plugin: ds %s was not properly initialized.", ds_name);
1060 sstrncpy(vtmp->vlist.type, ceph_dset_types[type], sizeof(vtmp->vlist.type));
1061 sstrncpy(vtmp->vlist.type_instance, ds_name, sizeof(vtmp->vlist.type_instance));
1062 vtmp->vlist.values = &uv;
1063 vtmp->vlist.values_len = 1;
1065 vtmp->index = (vtmp->index + 1);
1066 plugin_dispatch_values(&vtmp->vlist);
1071 static int cconn_connect(struct cconn *io)
1073 struct sockaddr_un address;
1075 if(io->state != CSTATE_UNCONNECTED)
1077 ERROR("ceph plugin: cconn_connect: io->state != CSTATE_UNCONNECTED");
1080 fd = socket(PF_UNIX, SOCK_STREAM, 0);
1084 ERROR("ceph plugin: cconn_connect: socket(PF_UNIX, SOCK_STREAM, 0) "
1085 "failed: error %d", err);
1088 memset(&address, 0, sizeof(struct sockaddr_un));
1089 address.sun_family = AF_UNIX;
1090 snprintf(address.sun_path, sizeof(address.sun_path), "%s",
1093 connect(fd, (struct sockaddr *) &address, sizeof(struct sockaddr_un)));
1096 ERROR("ceph plugin: cconn_connect: connect(%d) failed: error %d",
1102 flags = fcntl(fd, F_GETFL, 0);
1103 if(fcntl(fd, F_SETFL, flags | O_NONBLOCK) != 0)
1106 ERROR("ceph plugin: cconn_connect: fcntl(%d, O_NONBLOCK) error %d",
1112 io->state = CSTATE_WRITE_REQUEST;
1119 static void cconn_close(struct cconn *io)
1121 io->state = CSTATE_UNCONNECTED;
1125 RETRY_ON_EINTR(res, close(io->asok));
1134 /* Process incoming JSON counter data */
1136 cconn_process_data(struct cconn *io, yajl_struct *yajl, yajl_handle hand)
1139 struct values_tmp *vtmp = calloc(1, sizeof(struct values_tmp) * 1);
1145 vtmp->vlist = (value_list_t)VALUE_LIST_INIT;
1146 sstrncpy(vtmp->vlist.host, hostname_g, sizeof(vtmp->vlist.host));
1147 sstrncpy(vtmp->vlist.plugin, "ceph", sizeof(vtmp->vlist.plugin));
1148 sstrncpy(vtmp->vlist.plugin_instance, io->d->name, sizeof(vtmp->vlist.plugin_instance));
1151 vtmp->avgcount_exists = -1;
1152 vtmp->latency_index = 0;
1154 yajl->handler_arg = vtmp;
1155 ret = traverse_json(io->json, io->json_len, hand);
1161 * Initiate JSON parsing and print error if one occurs
1163 static int cconn_process_json(struct cconn *io)
1165 if((io->request_type != ASOK_REQ_DATA) &&
1166 (io->request_type != ASOK_REQ_SCHEMA))
1175 hand = yajl_alloc(&callbacks,
1177 /* alloc funcs = */ NULL,
1179 /* alloc funcs = */ NULL, NULL,
1181 /* context = */ (void *)(&io->yajl));
1185 ERROR ("ceph plugin: yajl_alloc failed.");
1191 switch(io->request_type)
1194 io->yajl.handler = node_handler_fetch_data;
1195 result = cconn_process_data(io, &io->yajl, hand);
1197 case ASOK_REQ_SCHEMA:
1198 //init daemon specific variables
1200 io->d->last_idx = 0;
1201 io->d->last_poll_data = NULL;
1202 io->yajl.handler = node_handler_define_schema;
1203 io->yajl.handler_arg = io->d;
1204 result = traverse_json(io->json, io->json_len, hand);
1214 status = yajl_complete_parse(hand);
1216 status = yajl_parse_complete(hand);
1219 if (status != yajl_status_ok)
1221 unsigned char *errmsg = yajl_get_error (hand, /* verbose = */ 0,
1222 /* jsonText = */ NULL, /* jsonTextLen = */ 0);
1223 ERROR ("ceph plugin: yajl_parse_complete failed: %s",
1225 yajl_free_error (hand, errmsg);
1235 static int cconn_validate_revents(struct cconn *io, int revents)
1237 if(revents & POLLERR)
1239 ERROR("ceph plugin: cconn_validate_revents(name=%s): got POLLERR",
1245 case CSTATE_WRITE_REQUEST:
1246 return (revents & POLLOUT) ? 0 : -EINVAL;
1247 case CSTATE_READ_VERSION:
1248 case CSTATE_READ_AMT:
1249 case CSTATE_READ_JSON:
1250 return (revents & POLLIN) ? 0 : -EINVAL;
1252 ERROR("ceph plugin: cconn_validate_revents(name=%s) got to "
1253 "illegal state on line %d", io->d->name, __LINE__);
1258 /** Handle a network event for a connection */
1259 static int cconn_handle_event(struct cconn *io)
1264 case CSTATE_UNCONNECTED:
1265 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1266 "state on line %d", io->d->name, __LINE__);
1269 case CSTATE_WRITE_REQUEST:
1272 snprintf(cmd, sizeof(cmd), "%s%d%s", "{ \"prefix\": \"",
1273 io->request_type, "\" }\n");
1274 size_t cmd_len = strlen(cmd);
1276 write(io->asok, ((char*)&cmd) + io->amt, cmd_len - io->amt));
1277 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,amt=%d,ret=%d)",
1278 io->d->name, io->state, io->amt, ret);
1284 if(io->amt >= cmd_len)
1287 switch (io->request_type)
1289 case ASOK_REQ_VERSION:
1290 io->state = CSTATE_READ_VERSION;
1293 io->state = CSTATE_READ_AMT;
1299 case CSTATE_READ_VERSION:
1302 read(io->asok, ((char*)(&io->d->version)) + io->amt,
1303 sizeof(io->d->version) - io->amt));
1304 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1305 io->d->name, io->state, ret);
1311 if(io->amt >= sizeof(io->d->version))
1313 io->d->version = ntohl(io->d->version);
1314 if(io->d->version != 1)
1316 ERROR("ceph plugin: cconn_handle_event(name=%s) not "
1317 "expecting version %d!", io->d->name, io->d->version);
1320 DEBUG("ceph plugin: cconn_handle_event(name=%s): identified as "
1321 "version %d", io->d->name, io->d->version);
1324 io->request_type = ASOK_REQ_SCHEMA;
1328 case CSTATE_READ_AMT:
1331 read(io->asok, ((char*)(&io->json_len)) + io->amt,
1332 sizeof(io->json_len) - io->amt));
1333 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1334 io->d->name, io->state, ret);
1340 if(io->amt >= sizeof(io->json_len))
1342 io->json_len = ntohl(io->json_len);
1344 io->state = CSTATE_READ_JSON;
1345 io->json = calloc(1, io->json_len + 1);
1348 ERROR("ceph plugin: error callocing io->json");
1354 case CSTATE_READ_JSON:
1357 read(io->asok, io->json + io->amt, io->json_len - io->amt));
1358 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1359 io->d->name, io->state, ret);
1365 if(io->amt >= io->json_len)
1367 ret = cconn_process_json(io);
1373 io->request_type = ASOK_REQ_NONE;
1378 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1379 "state on line %d", io->d->name, __LINE__);
1384 static int cconn_prepare(struct cconn *io, struct pollfd* fds)
1387 if(io->request_type == ASOK_REQ_NONE)
1389 /* The request has already been serviced. */
1392 else if((io->request_type == ASOK_REQ_DATA) && (io->d->ds_num == 0))
1394 /* If there are no counters to report on, don't bother
1401 case CSTATE_UNCONNECTED:
1402 ret = cconn_connect(io);
1412 fds->events = POLLOUT;
1414 case CSTATE_WRITE_REQUEST:
1416 fds->events = POLLOUT;
1418 case CSTATE_READ_VERSION:
1419 case CSTATE_READ_AMT:
1420 case CSTATE_READ_JSON:
1422 fds->events = POLLIN;
1425 ERROR("ceph plugin: cconn_prepare(name=%s) got to illegal state "
1426 "on line %d", io->d->name, __LINE__);
1431 /** Returns the difference between two struct timevals in milliseconds.
1432 * On overflow, we return max/min int.
1434 static int milli_diff(const struct timeval *t1, const struct timeval *t2)
1437 int sec_diff = t1->tv_sec - t2->tv_sec;
1438 int usec_diff = t1->tv_usec - t2->tv_usec;
1439 ret = usec_diff / 1000;
1440 ret += (sec_diff * 1000);
1441 return (ret > INT_MAX) ? INT_MAX : ((ret < INT_MIN) ? INT_MIN : (int)ret);
1444 /** This handles the actual network I/O to talk to the Ceph daemons.
1446 static int cconn_main_loop(uint32_t request_type)
1448 int i, ret, some_unreachable = 0;
1449 struct timeval end_tv;
1450 struct cconn io_array[g_num_daemons];
1452 DEBUG("ceph plugin: entering cconn_main_loop(request_type = %d)", request_type);
1454 /* create cconn array */
1455 memset(io_array, 0, sizeof(io_array));
1456 for(i = 0; i < g_num_daemons; ++i)
1458 io_array[i].d = g_daemons[i];
1459 io_array[i].request_type = request_type;
1460 io_array[i].state = CSTATE_UNCONNECTED;
1463 /** Calculate the time at which we should give up */
1464 gettimeofday(&end_tv, NULL);
1465 end_tv.tv_sec += CEPH_TIMEOUT_INTERVAL;
1471 struct cconn *polled_io_array[g_num_daemons];
1472 struct pollfd fds[g_num_daemons];
1473 memset(fds, 0, sizeof(fds));
1475 for(i = 0; i < g_num_daemons; ++i)
1477 struct cconn *io = io_array + i;
1478 ret = cconn_prepare(io, fds + nfds);
1481 WARNING("ceph plugin: cconn_prepare(name=%s,i=%d,st=%d)=%d",
1482 io->d->name, i, io->state, ret);
1484 io->request_type = ASOK_REQ_NONE;
1485 some_unreachable = 1;
1489 polled_io_array[nfds++] = io_array + i;
1498 gettimeofday(&tv, NULL);
1499 diff = milli_diff(&end_tv, &tv);
1504 WARNING("ceph plugin: cconn_main_loop: timed out.");
1507 RETRY_ON_EINTR(ret, poll(fds, nfds, diff));
1510 ERROR("ceph plugin: poll(2) error: %d", ret);
1513 for(i = 0; i < nfds; ++i)
1515 struct cconn *io = polled_io_array[i];
1516 int revents = fds[i].revents;
1521 else if(cconn_validate_revents(io, revents))
1523 WARNING("ceph plugin: cconn(name=%s,i=%d,st=%d): "
1524 "revents validation error: "
1525 "revents=0x%08x", io->d->name, i, io->state, revents);
1527 io->request_type = ASOK_REQ_NONE;
1528 some_unreachable = 1;
1532 int ret = cconn_handle_event(io);
1535 WARNING("ceph plugin: cconn_handle_event(name=%s,"
1536 "i=%d,st=%d): error %d", io->d->name, i, io->state, ret);
1538 io->request_type = ASOK_REQ_NONE;
1539 some_unreachable = 1;
1544 done: for(i = 0; i < g_num_daemons; ++i)
1546 cconn_close(io_array + i);
1548 if(some_unreachable)
1550 DEBUG("ceph plugin: cconn_main_loop: some Ceph daemons were unreachable.");
1554 DEBUG("ceph plugin: cconn_main_loop: reached all Ceph daemons :)");
1559 static int ceph_read(void)
1561 return cconn_main_loop(ASOK_REQ_DATA);
1564 /******* lifecycle *******/
1565 static int ceph_init(void)
1568 ceph_daemons_print();
1570 ret = cconn_main_loop(ASOK_REQ_VERSION);
1572 return (ret) ? ret : 0;
1575 static int ceph_shutdown(void)
1578 for(i = 0; i < g_num_daemons; ++i)
1580 ceph_daemon_free(g_daemons[i]);
1585 DEBUG("ceph plugin: finished ceph_shutdown");
1589 void module_register(void)
1591 plugin_register_complex_config("ceph", ceph_config);
1592 plugin_register_init("ceph", ceph_init);
1593 plugin_register_read("ceph", ceph_read);
1594 plugin_register_shutdown("ceph", ceph_shutdown);
1596 /* vim: set sw=4 sts=4 et : */