Allow netdata plugins to expose functions for querying more information about specific charts (#13720)

* function renames and code cleanup in popen.c; no actual code changes

* netdata popen() now opens both child process stdin and stdout and returns FILE * for both

* pass both input and output to parser structures

* updated rrdset to call custom functions

* RRDSET FUNCTION leading calls for both sync and async operation

* put RRDSET functions to a separate file

* added format and timeout at function definition

* support for synchronous (internal plugins) and asynchronous (external plugins and children) functions

* /api/v1/function endpoint

* functions are now attached to the host and there is a dictionary view per chart

* functions implemented at plugins.d

* remove the defer until keyword hook from plugins.d when it is done

* stream sender implementation of functions

* sanitization of all functions so that certain characters are only allowed

* strictier sanitization

* common max size

* 1st working plugins.d example

* always init inflight dictionary

* properly destroy dictionaries to avoid parallel insertion of items

* add more debugging on disconnection reasons

* add more debugging on disconnection reasons again

* streaming receiver respects newlines

* dont use the same fp for both streaming receive and send

* dont free dbengine memory with internal checks

* make sender proceed in the buffer

* added timing info and garbage collection at plugins.d

* added info about routing nodes

* added info about routing nodes with delay

* added more info about delays

* added more info about delays again

* signal sending thread to wake up

* streaming version labeling and commented code to support capabilities

* added functions to /api/v1/data, /api/v1/charts, /api/v1/chart, /api/v1/info

* redirect top output to stdout

* address coverity findings

* fix resource leaks of popen

* log attempts to connect to individual destinations

* better messages

* properly parse destinations

* try to find a function from the most matching to the least matching

* log added streaming destinations

* rotate destinations bypassing a node in the middle that does not accept our connection

* break the loops properly

* use typedef to define callbacks

* capabilities negotiation during streaming

* functions exposed upstream based on capabilities; compression disabled per node persisting reconnects; always try to connect with all capabilities

* restore functionality to lookup functions

* better logging of capabilities

* remove old versions from capabilities when a newer version is there

* fix formatting

* optimization for plugins.d rrdlabels to avoid creating and destructing dictionaries all the time

* delayed health initialization for rrddim and rrdset

* cleanup health initialization

* fix for popen() not returning the right value

* add health worker jobs for initializing rrdset and rrddim

* added content type support for functions; apps.plugin permanent function to display all the processes

* fixes for functions parameters parsing in apps.plugin

* fix for process matching in apps.plugiin

* first working function for apps.plugin

* Dashboard ACL is disabled for functions; Function errors are all in JSON format

* apps.plugin function processes returns json table

* use json_escape_string() to escape message

* fix formatting

* apps.plugin exposes all its metrics to function processes

* fix json formatting when filtering out some rows

* reopen the internal pipe of rrdpush in case of errors

* misplaced statement

* do not use buffer->len

* support for GLOBAL functions (functions that are not linked to a chart

* added /api/v1/functions endpoint; removed format from the FUNCTIONS api;

* swagger documentation about the new api end points

* added plugins.d documentation about functions

* never re-close a file

* remove uncessesary ifdef

* fixed issues identified by codacy

* fix for null label value

* make edit-config copy-and-paste friendly

* Revert "make edit-config copy-and-paste friendly"

This reverts commit 54500c0e0a97f65a0c66c4d34e966f6a9056698e.

* reworked sender handshake to fix coverity findings

* timeout is zero, for both send_timeout() and recv_timeout()

* properly detect that parent closed the socket

* support caching of function responses; limit function response to 10MB; added protection from malformed function responses

* disabled excessive logging

* added units to apps.plugin function processes and normalized all values to be human readable

* shorter field names

* fixed issues reported

* fixed apps.plugin error response; tested that pluginsd can properly handle faulty responses

* use double linked list macros for double linked list management

* faster apps.plugin function printing by minimizing file operations

* added memory percentage

* fix compatibility issues with older compilers and FreeBSD

* rrdpush sender code cleanup; rrhost structure cleanup from sender flags and variables;

* fix letftover variable in ifdef

* apps.plugin: do not call detach from the thread; exit immediately when input is broken

* exclude AR charts from health

* flush cleaner; prefer sender output

* clarity

* do not fill the cbuffer if not connected

* fix

* dont enabled host->sender if streaming is not enabled; send host label updates to parent;

* functions are only available through ACLK

* Prepared statement reports only in dev mode

* fix AR chart detection

* fix for streaming not being enabling itself

* more cleanup of sender and receiver structures

* moved read-only flags and configuration options to rrdhost->options

* fixed merge with master

* fix for incomplete rename

* prevent service thread from working on charts that are being collected

Co-authored-by: Stelios Fragkakis <52996999+stelfrag@users.noreply.github.com>

1 files changed, 53 insertions, 26 deletions

diff --git a/health/health.c b/health/health.c
index be8b50ebb3..05dd2ca010 100644
--- a/health/health.c
+++ b/health/health.c
@@ -2,6 +2,22 @@
 
 #include "health.h"
 
+#define WORKER_HEALTH_JOB_RRD_LOCK              0
+#define WORKER_HEALTH_JOB_HOST_LOCK             1
+#define WORKER_HEALTH_JOB_DB_QUERY              2
+#define WORKER_HEALTH_JOB_CALC_EVAL             3
+#define WORKER_HEALTH_JOB_WARNING_EVAL          4
+#define WORKER_HEALTH_JOB_CRITICAL_EVAL         5
+#define WORKER_HEALTH_JOB_ALARM_LOG_ENTRY       6
+#define WORKER_HEALTH_JOB_ALARM_LOG_PROCESS     7
+#define WORKER_HEALTH_JOB_DELAYED_INIT_RRDSET   8
+#define WORKER_HEALTH_JOB_DELAYED_INIT_RRDDIM   9
+
+#if WORKER_UTILIZATION_MAX_JOB_TYPES < 10
+#error WORKER_UTILIZATION_MAX_JOB_TYPES has to be at least 10
+#endif
+
+
 unsigned int default_health_enabled = 1;
 char *silencers_filename;
 
@@ -659,22 +675,48 @@ static int update_disabled_silenced(RRDHOST *host, RRDCALC *rc) {
         return 0;
 }
 
-// Create alarms for dimensions that have been added to charts
-// since the previous iteration.
-static void health_execute_pending_updates(RRDHOST *host) {
+static void health_execute_delayed_initializations(RRDHOST *host) {
     RRDSET *st;
 
-    if (!rrdhost_flag_check(host, RRDHOST_FLAG_PENDING_FOREACH_ALARMS))
-        return;
+    if (!rrdhost_flag_check(host, RRDHOST_FLAG_PENDING_HEALTH_INITIALIZATION)) return;
+    rrdhost_flag_clear(host, RRDHOST_FLAG_PENDING_HEALTH_INITIALIZATION);
 
     rrdset_foreach_reentrant(st, host) {
-        if(!rrdset_flag_check(st, RRDSET_FLAG_PENDING_FOREACH_ALARMS))
+        if(!rrdset_flag_check(st, RRDSET_FLAG_PENDING_HEALTH_INITIALIZATION)) continue;
+        rrdset_flag_clear(st, RRDSET_FLAG_PENDING_HEALTH_INITIALIZATION);
+
+        if(unlikely(rrdset_is_ar_chart(st)))
             continue;
 
+        worker_is_busy(WORKER_HEALTH_JOB_DELAYED_INIT_RRDSET);
+
+        if(!st->rrdfamily)
+            st->rrdfamily = rrdfamily_add_and_acquire(host, rrdset_family(st));
+
+        if(!st->rrdvars)
+            st->rrdvars = rrdvariables_create();
+
+        rrddimvar_index_init(st);
+
+        rrdsetvar_add_and_leave_released(st, "last_collected_t", RRDVAR_TYPE_TIME_T, &st->last_collected_time.tv_sec, RRDVAR_FLAG_NONE);
+        rrdsetvar_add_and_leave_released(st, "collected_total_raw", RRDVAR_TYPE_TOTAL, &st->last_collected_total, RRDVAR_FLAG_NONE);
+        rrdsetvar_add_and_leave_released(st, "green", RRDVAR_TYPE_CALCULATED, &st->green, RRDVAR_FLAG_NONE);
+        rrdsetvar_add_and_leave_released(st, "red", RRDVAR_TYPE_CALCULATED, &st->red, RRDVAR_FLAG_NONE);
+        rrdsetvar_add_and_leave_released(st, "update_every", RRDVAR_TYPE_INT, &st->update_every, RRDVAR_FLAG_NONE);
+
+        rrdcalc_link_matching_alerts_to_rrdset(st);
+        rrdcalctemplate_link_matching_templates_to_rrdset(st);
+
         RRDDIM *rd;
         rrddim_foreach_read(rd, st) {
-            if(!rrddim_flag_check(rd, RRDDIM_FLAG_PENDING_FOREACH_ALARMS))
-                continue;
+            if(!rrddim_flag_check(rd, RRDDIM_FLAG_PENDING_HEALTH_INITIALIZATION)) continue;
+            rrddim_flag_clear(rd, RRDDIM_FLAG_PENDING_HEALTH_INITIALIZATION);
+
+            worker_is_busy(WORKER_HEALTH_JOB_DELAYED_INIT_RRDDIM);
+
+            rrddimvar_add_and_leave_released(rd, RRDVAR_TYPE_CALCULATED, NULL, NULL, &rd->last_stored_value, RRDVAR_FLAG_NONE);
+            rrddimvar_add_and_leave_released(rd, RRDVAR_TYPE_COLLECTED, NULL, "_raw", &rd->last_collected_value, RRDVAR_FLAG_NONE);
+            rrddimvar_add_and_leave_released(rd, RRDVAR_TYPE_TIME_T, NULL, "_last_collected_t", &rd->last_collected_time.tv_sec, RRDVAR_FLAG_NONE);
 
             RRDCALCTEMPLATE *rt;
             foreach_rrdcalctemplate_read(host, rt) {
@@ -685,14 +727,10 @@ static void health_execute_pending_updates(RRDHOST *host) {
                     rrdcalctemplate_check_rrddim_conditions_and_link(rt, st, rd, host);
             }
             foreach_rrdcalctemplate_done(rt);
-
-            rrddim_flag_clear(rd, RRDDIM_FLAG_PENDING_FOREACH_ALARMS);
         }
         rrddim_foreach_done(rd);
-        rrdset_flag_clear(st, RRDSET_FLAG_PENDING_FOREACH_ALARMS);
     }
     rrdset_foreach_done(st);
-    rrdhost_flag_clear(host, RRDHOST_FLAG_PENDING_FOREACH_ALARMS);
 }
 
 /**
@@ -705,19 +743,6 @@ static void health_execute_pending_updates(RRDHOST *host) {
  * @return It always returns NULL
  */
 
-#define WORKER_HEALTH_JOB_RRD_LOCK           0
-#define WORKER_HEALTH_JOB_HOST_LOCK          1
-#define WORKER_HEALTH_JOB_DB_QUERY           2
-#define WORKER_HEALTH_JOB_CALC_EVAL          3
-#define WORKER_HEALTH_JOB_WARNING_EVAL       4
-#define WORKER_HEALTH_JOB_CRITICAL_EVAL      5
-#define WORKER_HEALTH_JOB_ALARM_LOG_ENTRY    6
-#define WORKER_HEALTH_JOB_ALARM_LOG_PROCESS  7
-
-#if WORKER_UTILIZATION_MAX_JOB_TYPES < 8
-#error WORKER_UTILIZATION_MAX_JOB_TYPES has to be at least 8
-#endif
-
 void *health_main(void *ptr) {
     worker_register("HEALTH");
     worker_register_job_name(WORKER_HEALTH_JOB_RRD_LOCK, "rrd lock");
@@ -728,6 +753,8 @@ void *health_main(void *ptr) {
     worker_register_job_name(WORKER_HEALTH_JOB_CRITICAL_EVAL, "critical eval");
     worker_register_job_name(WORKER_HEALTH_JOB_ALARM_LOG_ENTRY, "alarm log entry");
     worker_register_job_name(WORKER_HEALTH_JOB_ALARM_LOG_PROCESS, "alarm log process");
+    worker_register_job_name(WORKER_HEALTH_JOB_DELAYED_INIT_RRDSET, "rrdset init");
+    worker_register_job_name(WORKER_HEALTH_JOB_DELAYED_INIT_RRDDIM, "rrddim init");
 
     netdata_thread_cleanup_push(health_main_cleanup, ptr);
 
@@ -809,7 +836,7 @@ void *health_main(void *ptr) {
             if(likely(!host->health_log_fp) && (loop == 1 || loop % cleanup_sql_every_loop == 0))
                 sql_health_alarm_log_cleanup(host);
 
-            health_execute_pending_updates(host);
+            health_execute_delayed_initializations(host);
 
             worker_is_busy(WORKER_HEALTH_JOB_HOST_LOCK);