Metrics framework checkpoint.

- Metrics API is much more similar to the Logging framework's API now.

- Filters define all output and metrics collection now.

- Initial attempt at thresholding and generating notices.
This commit is contained in:
Seth Hall 2011-08-09 11:12:02 -04:00
parent f36310dc0e
commit 9fa3bcffdd
7 changed files with 229 additions and 108 deletions

View file

@ -1,19 +0,0 @@
@load frameworks/metrics
redef enum Metrics::ID += {
CONNS_ORIGINATED,
CONNS_RESPONDED
};
event bro_init()
{
Metrics::configure(CONNS_ORIGINATED, [$aggregation_mask=24, $break_interval=5mins]);
Metrics::configure(CONNS_RESPONDED, [$aggregation_mask=24, $break_interval=5mins]);
}
event connection_established(c: connection)
{
Metrics::add_data(CONNS_ORIGINATED, [$host=c$id$orig_h], 1);
Metrics::add_data(CONNS_RESPONDED, [$host=c$id$resp_h], 1);
}

View file

@ -1,20 +0,0 @@
@load frameworks/metrics
redef enum Metrics::ID += {
HTTP_REQUESTS_BY_STATUS_CODE,
HTTP_REQUESTS_BY_HOST,
};
event bro_init()
{
Metrics::configure(HTTP_REQUESTS_BY_STATUS_CODE, [$aggregation_mask=24, $break_interval=10secs]);
Metrics::configure(HTTP_REQUESTS_BY_HOST, [$break_interval=10secs]);
}
event HTTP::log_http(rec: HTTP::Info)
{
if ( rec?$host )
Metrics::add_data(HTTP_REQUESTS_BY_HOST, [$index=rec$host], 1);
if ( rec?$status_code )
Metrics::add_data(HTTP_REQUESTS_BY_STATUS_CODE, [$host=rec$id$orig_h, $index=fmt("%d", rec$status_code)], 1);
}

View file

@ -1,30 +1,28 @@
##! This is the implementation of the metrics framework
##! This is the implementation of the metrics framework.
module Metrics;
export {
redef enum Log::ID += { METRICS };
type ID: enum {
ALL,
NOTHING,
};
const default_aggregation_mask = 24 &redef;
const default_break_interval = 5mins &redef;
## The default interval used for "breaking" metrics and writing the
## current value to the logging stream.
const default_break_interval = 15mins &redef;
# TODO: configure a metrics filter logging stream to log the current
# metrics configuration in case someone is looking through
# old logs and the configuration has changed since then.
type Filter: record {
name: ID &optional;
## Global mask by which you'd like to aggregate traffic.
aggregation_mask: count &optional;
## This is essentially applying names to various subnets.
aggregation_table: table[subnet] of string &optional;
break_interval: interval &default=default_break_interval;
type Info: record {
ts: time &log;
metric_id: ID &log;
filter_name: string &log;
agg_subnet: string &log &optional;
index: string &log &optional;
value: count &log;
};
type Index: record {
type Entry: record {
## Host is the value to which this metric applies.
host: addr &optional;
@ -36,36 +34,71 @@ export {
## metric since multiple IP addresses could respond for the same Host
## header value.
index: string &default="";
## The value by which the counter should be increased in each filter
## where this entry is accepted.
increment: count &default=1;
};
type Info: record {
ts: time &log;
name: ID &log;
index: string &log &optional;
agg_subnet: string &log &optional;
value: count &log;
# TODO: configure a metrics filter logging stream to log the current
# metrics configuration in case someone is looking through
# old logs and the configuration has changed since then.
type Filter: record {
## The :bro:type:`Metrics::ID` that this filter applies to.
id: ID &optional;
## The name for this filter so that multiple filters can be
## applied to a single metrics to get a different view of the same
## metric data being collected (different aggregation, break, etc).
name: string &default="default";
## A predicate so that you can decide per index if you would like
## to accept the data being inserted.
pred: function(entry: Entry): bool &optional;
## Global mask by which you'd like to aggregate traffic.
aggregation_mask: count &optional;
## This is essentially applying names to various subnets.
aggregation_table: table[subnet] of string &optional;
## The interval at which the metric should be "broken" and written
## to the logging stream.
break_interval: interval &default=default_break_interval;
## This determines if the result of this filter is sent to the metrics
## logging stream. One use for the logging framework is as an internal
## thresholding and statistics gathering utility that is meant to
## never log but rather to generate notices and derive data.
log: bool &default=T;
## A straight threshold for generating a notice.
notice_threshold: count &optional;
## A series of thresholds at which to generate notices.
notice_thresholds: vector of count &optional;
## If this and a $notice_threshold value are set, this notice type
## will be generated by the metrics framework.
note: Notice::Type &optional;
};
global add_filter: function(name: ID, filter: Filter);
global add_data: function(name: ID, index: Index, increment: count);
global add_filter: function(id: ID, filter: Filter);
global add_data: function(id: ID, entry: Entry);
global log_metrics: event(rec: Info);
}
global metric_filters: table[ID] of Filter = table();
global metric_filters: table[ID] of vector of Filter = table();
type MetricIndex: table[string] of count &default=0;
type MetricTable: table[string] of MetricIndex;
global store: table[ID] of MetricTable = table();
# This is indexed by metric ID and stream filter name.
global store: table[ID, string] of MetricTable = table();
event bro_init()
# This stores the current threshold index for filters using the
# $notice_thresholds element.
global thresholds: table[string] of count = {} &default=0;
event bro_init() &priority=5
{
Log::create_stream(METRICS, [$columns=Info, $ev=log_metrics]);
}
function reset(name: ID)
function reset(filter: Filter)
{
store[name] = table();
store[filter$id, filter$name] = table();
}
event log_it(filter: Filter)
@ -73,70 +106,128 @@ event log_it(filter: Filter)
# If this node is the manager in a cluster, this needs to request values
# for this metric from all of the workers.
local id = filter$id;
local name = filter$name;
for ( agg_subnet in store[name] )
for ( agg_subnet in store[id, name] )
{
local metric_values = store[name][agg_subnet];
local metric_values = store[id, name][agg_subnet];
for ( index in metric_values )
{
local val = metric_values[index];
local m: Info = [$ts=network_time(),
$name=name,
$agg_subnet=fmt("%s", agg_subnet),
$index=index,
$metric_id=id,
$filter_name=name,
$agg_subnet=fmt("%s", agg_subnet),
$index=index,
$value=val];
if ( filter?$notice_threshold &&
m$value >= filter$notice_threshold )
{
print m;
NOTICE([$note=filter$note,
$msg=fmt("Metrics threshold crossed by %s %d/%d", m$agg_subnet, m$value, filter$notice_threshold),
$n=m$value]);
}
else if ( filter?$notice_thresholds &&
m$value >= filter$notice_thresholds[thresholds[cat(id,name)]] )
{
# TODO: implement this
}
# If there wasn't an index, remove the field.
if ( index == "" )
delete m$index;
# If there wasn't an aggregation subnet, remove the field.
if ( agg_subnet == "" )
delete m$agg_subnet;
Log::write(METRICS, m);
}
}
reset(name);
reset(filter);
schedule filter$break_interval { log_it(filter) };
}
function add_filter(name: ID, filter: Filter)
function add_filter(id: ID, filter: Filter)
{
if ( filter?$aggregation_table && filter?$aggregation_mask )
{
print "INVALID Metric filter: Defined $aggregation_table and $aggregation_mask.";
return;
}
if ( [id, filter$name] in store )
{
print fmt("INVALID Metric filter: Filter with name \"%s\" already exists.", filter$name);
return;
}
if ( filter?$notice_threshold && filter?$notice_thresholds )
{
print "INVALID Metric filter: Defined both $notice_threshold and $notice_thresholds";
return;
}
filter$name = name;
metric_filters[name] = filter;
store[name] = table();
if ( ! filter?$id )
filter$id = id;
if ( id !in metric_filters )
metric_filters[id] = vector();
metric_filters[id][|metric_filters[id]|] = filter;
store[id, filter$name] = table();
# Only do this on the manager if in a cluster.
schedule filter$break_interval { log_it(filter) };
}
function add_data(name: ID, index: Index, increment: count)
function add_data(id: ID, entry: Entry)
{
local conf = metric_filters[name];
local agg_subnet = "";
if ( index?$host )
if ( id !in metric_filters )
return;
local filters = metric_filters[id];
# Add the data to any of the defined filters.
for ( filter_id in filters )
{
if ( conf?$aggregation_mask )
local filter = filters[filter_id];
# If this filter has a predicate, run the predicate and skip this
# entry if the predicate return false.
if ( filter?$pred &&
! filter$pred(entry) )
next;
local agg_subnet = "";
local filt_store = store[id, filter$name];
if ( entry?$host )
{
local agg_mask = conf$aggregation_mask;
agg_subnet = fmt("%s", mask_addr(index$host, agg_mask));
if ( filter?$aggregation_mask )
{
local agg_mask = filter$aggregation_mask;
agg_subnet = fmt("%s", mask_addr(entry$host, agg_mask));
}
else if ( filter?$aggregation_table )
{
agg_subnet = fmt("%s", filter$aggregation_table[entry$host]);
# if an aggregation table is being used and the value isn't
# in the table, that means we aren't interested in it.
if ( agg_subnet == "" )
next;
}
else
agg_subnet = fmt("%s", entry$host);
}
else if ( conf?$aggregation_table )
agg_subnet = fmt("%s", conf$aggregation_table[index$host]);
else
agg_subnet = fmt("%s", index$host);
if ( agg_subnet !in filt_store )
filt_store[agg_subnet] = table();
local fs = filt_store[agg_subnet];
if ( entry$index !in fs )
fs[entry$index] = 0;
fs[entry$index] = fs[entry$index] + entry$increment;
}
if ( agg_subnet !in store[name] )
store[name][agg_subnet] = table();
if ( index$index !in store[name][agg_subnet] )
store[name][agg_subnet][index$index] = 0;
store[name][agg_subnet][index$index] = store[name][agg_subnet][index$index] + increment;
}