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>
53 #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 */
99 DSET_TYPE_UNFOUND = 1000
102 /** Valid types for ceph defined in types.db */
103 static const char * const ceph_dset_types [CEPH_DSET_TYPES_NUM] =
104 {"ceph_latency", "ceph_bytes", "ceph_rate"};
106 /******* ceph_daemon *******/
109 /** Version of the admin_socket interface */
112 char name[DATA_MAX_NAME_LEN];
114 /** Path to the socket that we use to talk to the ceph daemon */
115 char asok_path[UNIX_DOMAIN_SOCK_PATH_MAX];
117 /** Number of counters */
119 /** Track ds types */
121 /** Track ds names to match with types */
125 * Keep track of last data for latency values so we can calculate rate
128 struct last_data **last_poll_data;
129 /** index of last poll data */
133 /******* JSON parsing *******/
134 typedef int (*node_handler_t)(void *, const char*, const char*);
136 /** Track state and handler while parsing JSON */
139 node_handler_t handler;
143 char *stack[YAJL_MAX_DEPTH];
146 typedef struct yajl_struct yajl_struct;
148 enum perfcounter_type_d
150 PERFCOUNTER_LATENCY = 0x4, PERFCOUNTER_DERIVE = 0x8,
153 /** Give user option to use default (long run = since daemon started) avg */
154 static int long_run_latency_avg = 0;
157 * Give user option to use default type for special cases -
158 * filestore.journal_wr_bytes is currently only metric here. Ceph reports the
159 * type as a sum/count pair and will calculate it the same as a latency value.
160 * All other "bytes" metrics (excluding the used/capacity bytes for the OSD)
161 * use the DERIVE type. Unless user specifies to use given type, convert this
162 * metric to use DERIVE.
164 static int convert_special_metrics = 1;
166 /** Array of daemons to monitor */
167 static struct ceph_daemon **g_daemons = NULL;
169 /** Number of elements in g_daemons */
170 static size_t g_num_daemons = 0;
173 * A set of data that we build up in memory while parsing the JSON.
177 /** ceph daemon we are processing data for*/
178 struct ceph_daemon *d;
179 /** track avgcount across counters for avgcount/sum latency pairs */
181 /** current index of counters - used to get type of counter */
183 /** do we already have an avgcount for latency pair */
186 * similar to index, but current index of latency type counters -
187 * used to get last poll data of counter
191 * values list - maintain across counters since
192 * host/plugin/plugin instance are always the same
198 * A set of count/sum pairs to keep track of latency types and get difference
199 * between this poll data and last poll data.
203 char ds_name[DATA_MAX_NAME_LEN];
208 /******* network I/O *******/
211 CSTATE_UNCONNECTED = 0,
212 CSTATE_WRITE_REQUEST,
220 ASOK_REQ_VERSION = 0,
223 ASOK_REQ_NONE = 1000,
228 /** The Ceph daemon that we're talking to */
229 struct ceph_daemon *d;
232 uint32_t request_type;
234 /** The connection state */
237 /** The socket we use to talk to this daemon */
240 /** The amount of data remaining to read / write. */
243 /** Length of the JSON to read */
246 /** Buffer containing JSON data */
249 /** Keep data important to yajl processing */
250 struct yajl_struct yajl;
253 static int ceph_cb_null(void *ctx)
255 return CEPH_CB_CONTINUE;
258 static int ceph_cb_boolean(void *ctx, int bool_val)
260 return CEPH_CB_CONTINUE;
263 #define BUFFER_ADD(dest, src) do { \
264 size_t dest_size = sizeof (dest); \
265 size_t dest_len = strlen (dest); \
266 if (dest_size > dest_len) { \
267 sstrncpy ((dest) + dest_len, (src), dest_size - dest_len); \
269 (dest)[dest_size - 1] = 0; \
273 ceph_cb_number(void *ctx, const char *number_val, yajl_len_t number_len)
275 yajl_struct *state = (yajl_struct*) ctx;
276 char buffer[number_len+1];
277 char key[2 * DATA_MAX_NAME_LEN] = { 0 };
278 _Bool latency_type = 0;
281 memcpy(buffer, number_val, number_len);
282 buffer[sizeof(buffer) - 1] = '\0';
284 for (size_t i = 0; i < state->depth; i++)
286 if (state->stack[i] == NULL)
289 if (strlen (key) != 0)
290 BUFFER_ADD (key, ".");
291 BUFFER_ADD (key, state->stack[i]);
294 /* Special case for latency metrics. */
295 if ((strcmp ("avgcount", state->key) == 0)
296 || (strcmp ("sum", state->key) == 0))
300 /* Super-special case for filestore.journal_wr_bytes.avgcount: For
301 * some reason, Ceph schema encodes this as a count/sum pair while all
302 * other "Bytes" data (excluding used/capacity bytes for OSD space) uses
303 * a single "Derive" type. To spare further confusion, keep this KPI as
304 * the same type of other "Bytes". Instead of keeping an "average" or
305 * "rate", use the "sum" in the pair and assign that to the derive
307 if (convert_special_metrics && (state->depth >= 2)
308 && (strcmp("filestore", state->stack[state->depth - 2]) == 0)
309 && (strcmp("journal_wr_bytes", state->stack[state->depth - 1]) == 0)
310 && (strcmp("avgcount", state->key) == 0))
312 DEBUG("ceph plugin: Skipping avgcount for filestore.JournalWrBytes");
313 return CEPH_CB_CONTINUE;
316 else /* not a latency type */
318 BUFFER_ADD (key, ".");
319 BUFFER_ADD (key, state->key);
322 status = state->handler(state->handler_arg, buffer, key);
323 if((status == RETRY_AVGCOUNT) && latency_type)
325 /* Add previously skipped part of the key, either "avgcount" or "sum",
327 BUFFER_ADD (key, ".");
328 BUFFER_ADD (key, state->key);
330 status = state->handler(state->handler_arg, buffer, key);
335 ERROR("ceph plugin: JSON handler failed with status %d.", status);
336 return CEPH_CB_ABORT;
339 return CEPH_CB_CONTINUE;
342 static int ceph_cb_string(void *ctx, const unsigned char *string_val,
343 yajl_len_t string_len)
345 return CEPH_CB_CONTINUE;
348 static int ceph_cb_start_map(void *ctx)
350 yajl_struct *state = (yajl_struct*) ctx;
352 /* Push key to the stack */
353 if (state->depth == YAJL_MAX_DEPTH)
354 return CEPH_CB_ABORT;
356 state->stack[state->depth] = state->key;
360 return CEPH_CB_CONTINUE;
363 static int ceph_cb_end_map(void *ctx)
365 yajl_struct *state = (yajl_struct*) ctx;
367 /* Pop key from the stack */
368 if (state->depth == 0)
369 return CEPH_CB_ABORT;
373 state->key = state->stack[state->depth];
374 state->stack[state->depth] = NULL;
376 return CEPH_CB_CONTINUE;
380 ceph_cb_map_key(void *ctx, const unsigned char *key, yajl_len_t string_len)
382 yajl_struct *state = (yajl_struct*) ctx;
383 size_t sz = ((size_t) string_len) + 1;
386 state->key = malloc (sz);
387 if (state->key == NULL)
389 ERROR ("ceph plugin: malloc failed.");
390 return CEPH_CB_ABORT;
393 memmove (state->key, key, sz - 1);
394 state->key[sz - 1] = 0;
396 return CEPH_CB_CONTINUE;
399 static int ceph_cb_start_array(void *ctx)
401 return CEPH_CB_CONTINUE;
404 static int ceph_cb_end_array(void *ctx)
406 return CEPH_CB_CONTINUE;
409 static yajl_callbacks callbacks = {
423 static void ceph_daemon_print(const struct ceph_daemon *d)
425 DEBUG("ceph plugin: name=%s, asok_path=%s", d->name, d->asok_path);
428 static void ceph_daemons_print(void)
430 for(size_t i = 0; i < g_num_daemons; ++i)
432 ceph_daemon_print(g_daemons[i]);
436 static void ceph_daemon_free(struct ceph_daemon *d)
438 for(int i = 0; i < d->last_idx; i++)
440 sfree(d->last_poll_data[i]);
442 sfree(d->last_poll_data);
443 d->last_poll_data = NULL;
446 for(int i = 0; i < d->ds_num; i++)
448 sfree(d->ds_names[i]);
455 /* compact_ds_name removed the special characters ":", "_", "-" and "+" from the
456 * intput string. Characters following these special characters are capitalized.
457 * Trailing "+" and "-" characters are replaces with the strings "Plus" and
459 static int compact_ds_name (char *buffer, size_t buffer_size, char const *src)
464 size_t ptr_size = buffer_size;
465 _Bool append_plus = 0;
466 _Bool append_minus = 0;
468 if ((buffer == NULL) || (buffer_size <= strlen ("Minus")) || (src == NULL))
471 src_copy = strdup (src);
472 src_len = strlen(src);
474 /* Remove trailing "+" and "-". */
475 if (src_copy[src_len - 1] == '+')
479 src_copy[src_len] = 0;
481 else if (src_copy[src_len - 1] == '-')
485 src_copy[src_len] = 0;
488 /* Split at special chars, capitalize first character, append to buffer. */
489 char *dummy = src_copy;
491 char *save_ptr = NULL;
492 while ((token = strtok_r (dummy, ":_-+", &save_ptr)) != NULL)
498 token[0] = toupper ((int) token[0]);
500 assert (ptr_size > 1);
502 len = strlen (token);
507 assert (len < ptr_size);
509 sstrncpy (ptr, token, len + 1);
518 /* Append "Plus" or "Minus" if "+" or "-" has been stripped above. */
519 if (append_plus || append_minus)
521 char const *append = "Plus";
525 size_t offset = buffer_size - (strlen (append) + 1);
526 if (offset > strlen (buffer))
527 offset = strlen (buffer);
529 sstrncpy (buffer + offset, append, buffer_size - offset);
536 static _Bool has_suffix (char const *str, char const *suffix)
538 size_t str_len = strlen (str);
539 size_t suffix_len = strlen (suffix);
542 if (suffix_len > str_len)
544 offset = str_len - suffix_len;
546 if (strcmp (str + offset, suffix) == 0)
552 /* count_parts returns the number of elements a "foo.bar.baz" style key has. */
553 static size_t count_parts (char const *key)
555 size_t parts_num = 0;
557 for (const char *ptr = key; ptr != NULL; ptr = strchr (ptr + 1, '.'))
564 * Parse key to remove "type" if this is for schema and initiate compaction
566 static int parse_keys (char *buffer, size_t buffer_size, const char *key_str)
568 char tmp[2 * buffer_size];
570 if (buffer == NULL || buffer_size == 0 || key_str == NULL || strlen (key_str) == 0)
573 if ((count_parts (key_str) > 2) && has_suffix (key_str, ".type"))
575 /* strip ".type" suffix iff the key has more than two parts. */
576 size_t sz = strlen (key_str) - strlen (".type") + 1;
578 if (sz > sizeof (tmp))
580 sstrncpy (tmp, key_str, sz);
584 sstrncpy (tmp, key_str, sizeof (tmp));
587 return compact_ds_name (buffer, buffer_size, tmp);
591 * while parsing ceph admin socket schema, save counter name and type for later
594 static int ceph_daemon_add_ds_entry(struct ceph_daemon *d, const char *name,
598 char ds_name[DATA_MAX_NAME_LEN];
600 if(convert_special_metrics)
603 * Special case for filestore:JournalWrBytes. For some reason, Ceph
604 * schema encodes this as a count/sum pair while all other "Bytes" data
605 * (excluding used/capacity bytes for OSD space) uses a single "Derive"
606 * type. To spare further confusion, keep this KPI as the same type of
607 * other "Bytes". Instead of keeping an "average" or "rate", use the
608 * "sum" in the pair and assign that to the derive value.
610 if((strcmp(name,"filestore.journal_wr_bytes.type") == 0))
616 d->ds_names = realloc(d->ds_names, sizeof(char *) * (d->ds_num + 1));
622 d->ds_types = realloc(d->ds_types, sizeof(uint32_t) * (d->ds_num + 1));
628 d->ds_names[d->ds_num] = malloc(DATA_MAX_NAME_LEN);
629 if(!d->ds_names[d->ds_num])
634 type = (pc_type & PERFCOUNTER_DERIVE) ? DSET_RATE :
635 ((pc_type & PERFCOUNTER_LATENCY) ? DSET_LATENCY : DSET_BYTES);
636 d->ds_types[d->ds_num] = type;
638 if (parse_keys(ds_name, sizeof (ds_name), name))
643 sstrncpy(d->ds_names[d->ds_num], ds_name, DATA_MAX_NAME_LEN -1);
644 d->ds_num = (d->ds_num + 1);
649 /******* ceph_config *******/
650 static int cc_handle_str(struct oconfig_item_s *item, char *dest, int dest_len)
653 if(item->values_num != 1)
657 if(item->values[0].type != OCONFIG_TYPE_STRING)
661 val = item->values[0].value.string;
662 if(snprintf(dest, dest_len, "%s", val) > (dest_len - 1))
664 ERROR("ceph plugin: configuration parameter '%s' is too long.\n",
666 return -ENAMETOOLONG;
671 static int cc_handle_bool(struct oconfig_item_s *item, int *dest)
673 if(item->values_num != 1)
678 if(item->values[0].type != OCONFIG_TYPE_BOOLEAN)
683 *dest = (item->values[0].value.boolean) ? 1 : 0;
687 static int cc_add_daemon_config(oconfig_item_t *ci)
690 struct ceph_daemon *nd, cd = { 0 };
691 struct ceph_daemon **tmp;
693 if((ci->values_num != 1) || (ci->values[0].type != OCONFIG_TYPE_STRING))
695 WARNING("ceph plugin: `Daemon' blocks need exactly one string "
700 ret = cc_handle_str(ci, cd.name, DATA_MAX_NAME_LEN);
706 for(int i=0; i < ci->children_num; i++)
708 oconfig_item_t *child = ci->children + i;
710 if(strcasecmp("SocketPath", child->key) == 0)
712 ret = cc_handle_str(child, cd.asok_path, sizeof(cd.asok_path));
720 WARNING("ceph plugin: ignoring unknown option %s", child->key);
723 if(cd.name[0] == '\0')
725 ERROR("ceph plugin: you must configure a daemon name.\n");
728 else if(cd.asok_path[0] == '\0')
730 ERROR("ceph plugin(name=%s): you must configure an administrative "
731 "socket path.\n", cd.name);
734 else if(!((cd.asok_path[0] == '/') ||
735 (cd.asok_path[0] == '.' && cd.asok_path[1] == '/')))
737 ERROR("ceph plugin(name=%s): administrative socket paths must begin "
738 "with '/' or './' Can't parse: '%s'\n", cd.name, cd.asok_path);
742 tmp = realloc(g_daemons, (g_num_daemons+1) * sizeof(*g_daemons));
745 /* The positive return value here indicates that this is a
746 * runtime error, not a configuration error. */
751 nd = malloc(sizeof (*nd));
756 memcpy(nd, &cd, sizeof(*nd));
757 g_daemons[g_num_daemons] = nd;
762 static int ceph_config(oconfig_item_t *ci)
766 for(int i = 0; i < ci->children_num; ++i)
768 oconfig_item_t *child = ci->children + i;
769 if(strcasecmp("Daemon", child->key) == 0)
771 ret = cc_add_daemon_config(child);
774 ERROR("ceph plugin: Couldn't allocate memory");
779 //process other daemons and ignore this one
783 else if(strcasecmp("LongRunAvgLatency", child->key) == 0)
785 ret = cc_handle_bool(child, &long_run_latency_avg);
791 else if(strcasecmp("ConvertSpecialMetricTypes", child->key) == 0)
793 ret = cc_handle_bool(child, &convert_special_metrics);
801 WARNING("ceph plugin: ignoring unknown option %s", child->key);
808 * Parse JSON and get error message if present
811 traverse_json(const unsigned char *json, uint32_t json_len, yajl_handle hand)
813 yajl_status status = yajl_parse(hand, json, json_len);
818 case yajl_status_error:
819 msg = yajl_get_error(hand, /* verbose = */ 1,
820 /* jsonText = */ (unsigned char *) json,
821 (unsigned int) json_len);
822 ERROR ("ceph plugin: yajl_parse failed: %s", msg);
823 yajl_free_error(hand, msg);
825 case yajl_status_client_canceled:
833 * Add entry for each counter while parsing schema
836 node_handler_define_schema(void *arg, const char *val, const char *key)
838 struct ceph_daemon *d = (struct ceph_daemon *) arg;
841 return ceph_daemon_add_ds_entry(d, key, pc_type);
845 * Latency counter does not yet have an entry in last poll data - add it.
847 static int add_last(struct ceph_daemon *d, const char *ds_n, double cur_sum,
850 d->last_poll_data[d->last_idx] = malloc(sizeof (*d->last_poll_data[d->last_idx]));
851 if(!d->last_poll_data[d->last_idx])
855 sstrncpy(d->last_poll_data[d->last_idx]->ds_name,ds_n,
856 sizeof(d->last_poll_data[d->last_idx]->ds_name));
857 d->last_poll_data[d->last_idx]->last_sum = cur_sum;
858 d->last_poll_data[d->last_idx]->last_count = cur_count;
859 d->last_idx = (d->last_idx + 1);
864 * Update latency counter or add new entry if it doesn't exist
866 static int update_last(struct ceph_daemon *d, const char *ds_n, int index,
867 double cur_sum, uint64_t cur_count)
869 if((d->last_idx > index) && (strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0))
871 d->last_poll_data[index]->last_sum = cur_sum;
872 d->last_poll_data[index]->last_count = cur_count;
876 if(!d->last_poll_data)
878 d->last_poll_data = malloc(sizeof (*d->last_poll_data));
879 if(!d->last_poll_data)
886 struct last_data **tmp_last = realloc(d->last_poll_data,
887 ((d->last_idx+1) * sizeof(struct last_data *)));
892 d->last_poll_data = tmp_last;
894 return add_last(d, ds_n, cur_sum, cur_count);
898 * If using index guess failed (shouldn't happen, but possible if counters
899 * get rearranged), resort to searching for counter name
901 static int backup_search_for_last_avg(struct ceph_daemon *d, const char *ds_n)
903 for(int i = 0; i < d->last_idx; i++)
905 if(strcmp(d->last_poll_data[i]->ds_name, ds_n) == 0)
914 * Calculate average b/t current data and last poll data
915 * if last poll data exists
917 static double get_last_avg(struct ceph_daemon *d, const char *ds_n, int index,
918 double cur_sum, uint64_t cur_count)
920 double result = -1.1, sum_delt = 0.0;
921 uint64_t count_delt = 0;
923 if(d->last_idx > index)
925 if(strcmp(d->last_poll_data[index]->ds_name, ds_n) == 0)
929 //test previous index
930 else if((index > 0) && (strcmp(d->last_poll_data[index-1]->ds_name, ds_n) == 0))
932 tmp_index = (index - 1);
936 tmp_index = backup_search_for_last_avg(d, ds_n);
939 if((tmp_index > -1) && (cur_count > d->last_poll_data[tmp_index]->last_count))
941 sum_delt = (cur_sum - d->last_poll_data[tmp_index]->last_sum);
942 count_delt = (cur_count - d->last_poll_data[tmp_index]->last_count);
943 result = (sum_delt / count_delt);
951 if(update_last(d, ds_n, tmp_index, cur_sum, cur_count) == -ENOMEM)
959 * If using index guess failed, resort to searching for counter name
961 static uint32_t backup_search_for_type(struct ceph_daemon *d, char *ds_name)
963 for(int i = 0; i < d->ds_num; i++)
965 if(strcmp(d->ds_names[i], ds_name) == 0)
967 return d->ds_types[i];
970 return DSET_TYPE_UNFOUND;
974 * Process counter data and dispatch values
976 static int node_handler_fetch_data(void *arg, const char *val, const char *key)
981 struct values_tmp *vtmp = (struct values_tmp*) arg;
982 uint32_t type = DSET_TYPE_UNFOUND;
983 int index = vtmp->index;
985 char ds_name[DATA_MAX_NAME_LEN];
987 if (parse_keys (ds_name, sizeof (ds_name), key))
992 if(index >= vtmp->d->ds_num)
994 //don't overflow bounds of array
995 index = (vtmp->d->ds_num - 1);
999 * counters should remain in same order we parsed schema... we maintain the
1000 * index variable to keep track of current point in list of counters. first
1001 * use index to guess point in array for retrieving type. if that doesn't
1002 * work, use the old way to get the counter type
1004 if(strcmp(ds_name, vtmp->d->ds_names[index]) == 0)
1007 type = vtmp->d->ds_types[index];
1009 else if((index > 0) && (strcmp(ds_name, vtmp->d->ds_names[index-1]) == 0))
1012 type = vtmp->d->ds_types[index-1];
1015 if(type == DSET_TYPE_UNFOUND)
1017 //couldn't find right type by guessing, check the old way
1018 type = backup_search_for_type(vtmp->d, ds_name);
1024 if(vtmp->avgcount_exists == -1)
1026 sscanf(val, "%" PRIu64, &vtmp->avgcount);
1027 vtmp->avgcount_exists = 0;
1028 //return after saving avgcount - don't dispatch value
1029 //until latency calculation
1035 sscanf(val, "%lf", &sum);
1037 if(vtmp->avgcount == 0)
1042 /** User wants latency values as long run avg */
1043 if(long_run_latency_avg)
1045 result = (sum / vtmp->avgcount);
1049 result = get_last_avg(vtmp->d, ds_name, vtmp->latency_index, sum, vtmp->avgcount);
1050 if(result == -ENOMEM)
1057 vtmp->avgcount_exists = -1;
1058 vtmp->latency_index = (vtmp->latency_index + 1);
1062 sscanf(val, "%lf", &tmp_d);
1066 sscanf(val, "%" PRIu64, &tmp_u);
1069 case DSET_TYPE_UNFOUND:
1071 ERROR("ceph plugin: ds %s was not properly initialized.", ds_name);
1075 sstrncpy(vtmp->vlist.type, ceph_dset_types[type], sizeof(vtmp->vlist.type));
1076 sstrncpy(vtmp->vlist.type_instance, ds_name, sizeof(vtmp->vlist.type_instance));
1077 vtmp->vlist.values = &uv;
1078 vtmp->vlist.values_len = 1;
1080 vtmp->index = (vtmp->index + 1);
1081 plugin_dispatch_values(&vtmp->vlist);
1086 static int cconn_connect(struct cconn *io)
1088 struct sockaddr_un address = { 0 };
1090 if(io->state != CSTATE_UNCONNECTED)
1092 ERROR("ceph plugin: cconn_connect: io->state != CSTATE_UNCONNECTED");
1095 fd = socket(PF_UNIX, SOCK_STREAM, 0);
1099 ERROR("ceph plugin: cconn_connect: socket(PF_UNIX, SOCK_STREAM, 0) "
1100 "failed: error %d", err);
1103 address.sun_family = AF_UNIX;
1104 snprintf(address.sun_path, sizeof(address.sun_path), "%s",
1107 connect(fd, (struct sockaddr *) &address, sizeof(struct sockaddr_un)));
1110 ERROR("ceph plugin: cconn_connect: connect(%d) failed: error %d",
1116 flags = fcntl(fd, F_GETFL, 0);
1117 if(fcntl(fd, F_SETFL, flags | O_NONBLOCK) != 0)
1120 ERROR("ceph plugin: cconn_connect: fcntl(%d, O_NONBLOCK) error %d",
1126 io->state = CSTATE_WRITE_REQUEST;
1133 static void cconn_close(struct cconn *io)
1135 io->state = CSTATE_UNCONNECTED;
1139 RETRY_ON_EINTR(res, close(io->asok));
1148 /* Process incoming JSON counter data */
1150 cconn_process_data(struct cconn *io, yajl_struct *yajl, yajl_handle hand)
1153 struct values_tmp *vtmp = calloc(1, sizeof(struct values_tmp) * 1);
1159 vtmp->vlist = (value_list_t)VALUE_LIST_INIT;
1160 sstrncpy(vtmp->vlist.host, hostname_g, sizeof(vtmp->vlist.host));
1161 sstrncpy(vtmp->vlist.plugin, "ceph", sizeof(vtmp->vlist.plugin));
1162 sstrncpy(vtmp->vlist.plugin_instance, io->d->name, sizeof(vtmp->vlist.plugin_instance));
1165 vtmp->avgcount_exists = -1;
1166 vtmp->latency_index = 0;
1168 yajl->handler_arg = vtmp;
1169 ret = traverse_json(io->json, io->json_len, hand);
1175 * Initiate JSON parsing and print error if one occurs
1177 static int cconn_process_json(struct cconn *io)
1179 if((io->request_type != ASOK_REQ_DATA) &&
1180 (io->request_type != ASOK_REQ_SCHEMA))
1189 hand = yajl_alloc(&callbacks,
1191 /* alloc funcs = */ NULL,
1193 /* alloc funcs = */ NULL, NULL,
1195 /* context = */ (void *)(&io->yajl));
1199 ERROR ("ceph plugin: yajl_alloc failed.");
1205 switch(io->request_type)
1208 io->yajl.handler = node_handler_fetch_data;
1209 result = cconn_process_data(io, &io->yajl, hand);
1211 case ASOK_REQ_SCHEMA:
1212 //init daemon specific variables
1214 io->d->last_idx = 0;
1215 io->d->last_poll_data = NULL;
1216 io->yajl.handler = node_handler_define_schema;
1217 io->yajl.handler_arg = io->d;
1218 result = traverse_json(io->json, io->json_len, hand);
1228 status = yajl_complete_parse(hand);
1230 status = yajl_parse_complete(hand);
1233 if (status != yajl_status_ok)
1235 unsigned char *errmsg = yajl_get_error (hand, /* verbose = */ 0,
1236 /* jsonText = */ NULL, /* jsonTextLen = */ 0);
1237 ERROR ("ceph plugin: yajl_parse_complete failed: %s",
1239 yajl_free_error (hand, errmsg);
1249 static int cconn_validate_revents(struct cconn *io, int revents)
1251 if(revents & POLLERR)
1253 ERROR("ceph plugin: cconn_validate_revents(name=%s): got POLLERR",
1259 case CSTATE_WRITE_REQUEST:
1260 return (revents & POLLOUT) ? 0 : -EINVAL;
1261 case CSTATE_READ_VERSION:
1262 case CSTATE_READ_AMT:
1263 case CSTATE_READ_JSON:
1264 return (revents & POLLIN) ? 0 : -EINVAL;
1266 ERROR("ceph plugin: cconn_validate_revents(name=%s) got to "
1267 "illegal state on line %d", io->d->name, __LINE__);
1272 /** Handle a network event for a connection */
1273 static int cconn_handle_event(struct cconn *io)
1278 case CSTATE_UNCONNECTED:
1279 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1280 "state on line %d", io->d->name, __LINE__);
1283 case CSTATE_WRITE_REQUEST:
1286 snprintf(cmd, sizeof(cmd), "%s%d%s", "{ \"prefix\": \"",
1287 io->request_type, "\" }\n");
1288 size_t cmd_len = strlen(cmd);
1290 write(io->asok, ((char*)&cmd) + io->amt, cmd_len - io->amt));
1291 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,amt=%d,ret=%d)",
1292 io->d->name, io->state, io->amt, ret);
1298 if(io->amt >= cmd_len)
1301 switch (io->request_type)
1303 case ASOK_REQ_VERSION:
1304 io->state = CSTATE_READ_VERSION;
1307 io->state = CSTATE_READ_AMT;
1313 case CSTATE_READ_VERSION:
1316 read(io->asok, ((char*)(&io->d->version)) + io->amt,
1317 sizeof(io->d->version) - io->amt));
1318 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1319 io->d->name, io->state, ret);
1325 if(io->amt >= sizeof(io->d->version))
1327 io->d->version = ntohl(io->d->version);
1328 if(io->d->version != 1)
1330 ERROR("ceph plugin: cconn_handle_event(name=%s) not "
1331 "expecting version %d!", io->d->name, io->d->version);
1334 DEBUG("ceph plugin: cconn_handle_event(name=%s): identified as "
1335 "version %d", io->d->name, io->d->version);
1338 io->request_type = ASOK_REQ_SCHEMA;
1342 case CSTATE_READ_AMT:
1345 read(io->asok, ((char*)(&io->json_len)) + io->amt,
1346 sizeof(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 >= sizeof(io->json_len))
1356 io->json_len = ntohl(io->json_len);
1358 io->state = CSTATE_READ_JSON;
1359 io->json = calloc(1, io->json_len + 1);
1362 ERROR("ceph plugin: error callocing io->json");
1368 case CSTATE_READ_JSON:
1371 read(io->asok, io->json + io->amt, io->json_len - io->amt));
1372 DEBUG("ceph plugin: cconn_handle_event(name=%s,state=%d,ret=%d)",
1373 io->d->name, io->state, ret);
1379 if(io->amt >= io->json_len)
1381 ret = cconn_process_json(io);
1387 io->request_type = ASOK_REQ_NONE;
1392 ERROR("ceph plugin: cconn_handle_event(name=%s) got to illegal "
1393 "state on line %d", io->d->name, __LINE__);
1398 static int cconn_prepare(struct cconn *io, struct pollfd* fds)
1401 if(io->request_type == ASOK_REQ_NONE)
1403 /* The request has already been serviced. */
1406 else if((io->request_type == ASOK_REQ_DATA) && (io->d->ds_num == 0))
1408 /* If there are no counters to report on, don't bother
1415 case CSTATE_UNCONNECTED:
1416 ret = cconn_connect(io);
1426 fds->events = POLLOUT;
1428 case CSTATE_WRITE_REQUEST:
1430 fds->events = POLLOUT;
1432 case CSTATE_READ_VERSION:
1433 case CSTATE_READ_AMT:
1434 case CSTATE_READ_JSON:
1436 fds->events = POLLIN;
1439 ERROR("ceph plugin: cconn_prepare(name=%s) got to illegal state "
1440 "on line %d", io->d->name, __LINE__);
1445 /** Returns the difference between two struct timevals in milliseconds.
1446 * On overflow, we return max/min int.
1448 static int milli_diff(const struct timeval *t1, const struct timeval *t2)
1451 int sec_diff = t1->tv_sec - t2->tv_sec;
1452 int usec_diff = t1->tv_usec - t2->tv_usec;
1453 ret = usec_diff / 1000;
1454 ret += (sec_diff * 1000);
1455 return (ret > INT_MAX) ? INT_MAX : ((ret < INT_MIN) ? INT_MIN : (int)ret);
1458 /** This handles the actual network I/O to talk to the Ceph daemons.
1460 static int cconn_main_loop(uint32_t request_type)
1462 int ret, some_unreachable = 0;
1463 struct timeval end_tv;
1464 struct cconn io_array[g_num_daemons];
1466 DEBUG("ceph plugin: entering cconn_main_loop(request_type = %d)", request_type);
1468 /* create cconn array */
1469 memset(io_array, 0, sizeof(io_array));
1470 for(size_t i = 0; i < g_num_daemons; ++i)
1472 io_array[i].d = g_daemons[i];
1473 io_array[i].request_type = request_type;
1474 io_array[i].state = CSTATE_UNCONNECTED;
1477 /** Calculate the time at which we should give up */
1478 gettimeofday(&end_tv, NULL);
1479 end_tv.tv_sec += CEPH_TIMEOUT_INTERVAL;
1485 struct cconn *polled_io_array[g_num_daemons];
1486 struct pollfd fds[g_num_daemons];
1487 memset(fds, 0, sizeof(fds));
1489 for(size_t i = 0; i < g_num_daemons; ++i)
1491 struct cconn *io = io_array + i;
1492 ret = cconn_prepare(io, fds + nfds);
1495 WARNING("ceph plugin: cconn_prepare(name=%s,i=%zu,st=%d)=%d",
1496 io->d->name, i, io->state, ret);
1498 io->request_type = ASOK_REQ_NONE;
1499 some_unreachable = 1;
1503 polled_io_array[nfds++] = io_array + i;
1512 gettimeofday(&tv, NULL);
1513 diff = milli_diff(&end_tv, &tv);
1518 WARNING("ceph plugin: cconn_main_loop: timed out.");
1521 RETRY_ON_EINTR(ret, poll(fds, nfds, diff));
1524 ERROR("ceph plugin: poll(2) error: %d", ret);
1527 for(int i = 0; i < nfds; ++i)
1529 struct cconn *io = polled_io_array[i];
1530 int revents = fds[i].revents;
1536 else if(cconn_validate_revents(io, revents))
1538 WARNING("ceph plugin: cconn(name=%s,i=%d,st=%d): "
1539 "revents validation error: "
1540 "revents=0x%08x", io->d->name, i, io->state, revents);
1542 io->request_type = ASOK_REQ_NONE;
1543 some_unreachable = 1;
1547 ret = cconn_handle_event(io);
1550 WARNING("ceph plugin: cconn_handle_event(name=%s,"
1551 "i=%d,st=%d): error %d", io->d->name, i, io->state, ret);
1553 io->request_type = ASOK_REQ_NONE;
1554 some_unreachable = 1;
1559 done: for(size_t i = 0; i < g_num_daemons; ++i)
1561 cconn_close(io_array + i);
1563 if(some_unreachable)
1565 DEBUG("ceph plugin: cconn_main_loop: some Ceph daemons were unreachable.");
1569 DEBUG("ceph plugin: cconn_main_loop: reached all Ceph daemons :)");
1574 static int ceph_read(void)
1576 return cconn_main_loop(ASOK_REQ_DATA);
1579 /******* lifecycle *******/
1580 static int ceph_init(void)
1584 #if defined(HAVE_SYS_CAPABILITY_H) && defined(CAP_DAC_OVERRIDE)
1585 if (check_capability (CAP_DAC_OVERRIDE) != 0)
1588 WARNING ("ceph plugin: Running collectd as root, but the "
1589 "CAP_DAC_OVERRIDE capability is missing. The plugin's read "
1590 "function will probably fail. Is your init system dropping "
1593 WARNING ("ceph plugin: collectd doesn't have the CAP_DAC_OVERRIDE "
1594 "capability. If you don't want to run collectd as root, try running "
1595 "\"setcap cap_dac_override=ep\" on the collectd binary.");
1599 ceph_daemons_print();
1601 ret = cconn_main_loop(ASOK_REQ_VERSION);
1603 return (ret) ? ret : 0;
1606 static int ceph_shutdown(void)
1608 for(size_t i = 0; i < g_num_daemons; ++i)
1610 ceph_daemon_free(g_daemons[i]);
1615 DEBUG("ceph plugin: finished ceph_shutdown");
1619 void module_register(void)
1621 plugin_register_complex_config("ceph", ceph_config);
1622 plugin_register_init("ceph", ceph_init);
1623 plugin_register_read("ceph", ceph_read);
1624 plugin_register_shutdown("ceph", ceph_shutdown);
1626 /* vim: set sw=4 sts=4 et : */