2 commits

Author	SHA1	Message	Date
Costa Tsaousis	f466b8aef5	DYNCFG: dynamically configured alerts (#16779) ... * cleanup alerts * fix references * fix references * fix references * load alerts once and apply them to each node * simplify health_create_alarm_entry() * Compile without warnings with compiler flags: -Wall -Wextra -Wformat=2 -Wshadow -Wno-format-nonliteral -Winit-self * code re-organization and cleanup * generate patterns when applying prototypes; give unique dyncfg names to all alerts * eval expressions keep the source and the parsed_as as STRING pointers * renamed host to node in dyncfg ids * renamed host to node in dyncfg ids * add all cloud roles to the list of parsed X-Netdata-Role header and also default to member access level * working functionality * code re-organization: moved health event-loop to a new file, moved health globals to health.c * rrdcalctemplate is removed; alert_cfg is removed; foreach dimension is removed; RRDCALCs are now instanciated only when they are linked to RRDSETs * dyncfg alert prototypes initialization for alerts * health dyncfg split to separate file * cleanup not-needed code * normalize matches between parsing and json * also detect !* for disabled alerts * dyncfg capability disabled * Store alert config part1 * Add rrdlabels_common_count * wip health variables lookup without indexes * Improve rrdlabels_common_count by reusing rrdlabels_find_label_with_key_unsafe with an additional parameter * working variables with runtime lookup * working variables with runtime lookup * delete rrddimvar and rrdfamily index * remove rrdsetvar; now all variables are in RRDVARs inside hosts and charts * added /api/v1/variable that resolves a variable the same way alerts do * remove rrdcalc from eval * remove debug code * remove duplicate assignment * Fix memory leak * all alert variables are now handled by alert_variable_lookup() and EVAL is now independent of alerts * hide all internal structures of EVAL * Enable -Wformat flag Signed-off-by: Tasos Katsoulas <tasos@netdata.cloud> * Adjust binding for calculation, warning, critical * Remove unused macro * Update config hash id * use the right info and summary in alerts log * use synchronous queries for alerts * Handle cases when config_hash_id is missing from health_log * remove deadlock from health worker * parsing to json payload for health alert prototypes * cleaner parsing and avoiding memory leaks in case of duplicate members in json * fix left-over rename of function * Keep original lookup field to send to the cloud Cleanup / rename function to store config Remove unused DEFINEs, functions * Use ac->lookup * link jobs to the host when the template is registered; do not accept running a function without a host * full dyncfg support for health alerts, except action TEST * working dyncfg additions, updates, removals * fixed missing source, wrong status updates * add alerts by type, component, classification, recipient and module at the /api/v2/alerts endpoint * fix dyncfg unittest * rename functions * generalize the json-c parser macros and move them to libnetdata * report progress when enabling and disabling dyncfg templates * moved rrdcalc and rrdvar to health * update alarms * added schema for alerts; separated alert_action_options from rrdr_options; restructured the json payload for alerts * enable parsed json alerts; allow sending back accepted but disabled * added format_version for alerts payload; enables/disables status now is also inheritted by the status of the rules; fixed variable names in json output * remove the RRDHOST pointer from DYNCFG * Fix command field submitted to the cloud * do not send updates to creation requests, for DYNCFG jobs --------- Signed-off-by: Tasos Katsoulas <tasos@netdata.cloud> Co-authored-by: Stelios Fragkakis <52996999+stelfrag@users.noreply.github.com> Co-authored-by: Tasos Katsoulas <tasos@netdata.cloud> Co-authored-by: ilyam8 <ilya@netdata.cloud>	2024-01-23 20:20:41 +02:00
Costa Tsaousis	f2b250a1f5	dyncfg v2 (#16702) ... * split rrdfunctions streaming and progress * simplified internal inline functions API * split rrdfunctions inflight management * split rrd functions exporters * renames * base dyncfg structure * config pluginsd * intercept dyncfg function calls * loading and saving of dyncfg metadata and data * save metadata and payload to a single file; added code to update the plugins with jobs and saved configs * basic working unit test * added payload to functions execution * removed old dyncfg code that is not needed any more * more cleanup * cleanup sender for functions with payload * dyncfg functions are not exposed as functions * remaining work to avoid indexing the \0 terminating character in dictionary keys * added back old dyncfg plugins.d commands as noop, to allow plugins continue working * working api; working streaming; * updated plugins.d documentation * aclk and http api requests share the same header parsing logic * added source type internal * fixed crashes * added god mode for tests * fixes * fixed messages * save host machine guids to configs * cleaner manipulation of supported commands * the functions event loop for external plugins can now process dyncfg requests * unified internal and external plugins dyncfg API * Netdata serves schema requests from /etc/netdata/schema.d and /var/lib/netdata/conf.d/schema.d * cleanup and various fixes; fixed bug in previous dyncfg implementation on streaming that was sending the paylod in a way that allowed other streaming commands to be multiplexed * internals go to a separate header file * fix duplicate ACLK requests sent by aclk queue mechanism * use fstat instead of stat * working api * plugin actions renamed to create and delete; dyncfg files are removed only from user actions * prevent deadlock by using the react callback * fix for string_strndupz() * better dyncfg unittests * more tests at the unittests * properly detect dyncfg functions * hide config functions from the UI * tree response improvements * send the initial update with payload * determine tty using stdout, not stderr * changes to statuses, cleanup and the code to bring all business logic into interception * do not crash when the status is empty * functions now propagate the source of the requests to plugins * avoid warning about unused functions * in the count at items for attention, do not count the orphan entries * save source into dyncfg * make the list null terminated * fixed invalid comparison * prevent memory leak on duplicated headers; log x-forwarded-for * more unit tests * added dyncfg unittests into the default unittests * more unit tests and fixes * more unit tests and fixes * fix dictionary unittests * config functions require admin access	2024-01-11 16:56:45 +02:00