"Pop bottles." -- Birdman
Graphite clusters are pretty cool. Here are some primitive tools to help you manage your graphite clusters.
All of the tools support two common arguments; the path to a config file, and the name of the cluster. Using these tools alongside a config file that describes your graphite clusters you can build up scripts to manage your metrics. Some of the tools could easily be replaced with one-liners in shell, but exist here for convenience and readability. The goal is to provide fast, predictable utilities that can easily be composed into more advanced tooling.
Carbonate is available from Python official third party repository (aka PyPi) and as such can be installed via regular Python package managers. Note that you might have to install a python package manager (e.g. apt-get install python-setuptools on a ubuntu host)
pip install carbonate
Carbonate expects a configuration file that defines the clusters in your environment. The default config file is located at /opt/graphite/conf/carbonate.conf or can be provided on the command line. The default cluster is named 'main'. Both defaults can be overridden by setting in the environment CARBONATE_CONFIG and CARBONATE_CLUSTER respectively.
[main]
DESTINATIONS = 192.168.9.13:2004:carbon01, 192.168.9.15:2004:carbon02, 192.168.6.20:2004:carbon03
REPLICATION_FACTOR = 2
SSH_USER = carbon
[agg]
DESTINATIONS = 192.168.9.13:2004:carbon01, 192.168.9.15:2004:carbon02, 192.168.6.20:2004:carbon03
RELAY_METHOD = aggregated-consistent-hashing
REPLICATION_FACTOR = 2
SSH_USER = carbon
[fnv]
DESTINATIONS = 192.168.9.13:2004:ba603c36342304ed77953f84ac4d357b, 192.168.9.15:2004:5dd63865534f84899c6e5594dba6749a, 192.168.6.20:2004:866a18b81f2dc4649517a1df13e26f28
REPLICATION_FACTOR = 2
SSH_USER = carbonate
HASHING_TYPE = fnv1a_ch
You should take care to match the list of destination IPs or hostnames to the nodes in your cluster (i.e. it should match with routing configuretion of your carbon relay). Order is important because of how the consistent hash ring is created.
You can configure the relay method to be one of "consistent-hashing" or "aggregated-consistent-hashing". If omitted, "consistent-hashing" is used by default. Use of "aggregated-consistent-hashing" usually requires a rules file to be provided to relevant commands.
The replication factor should match the replication factor for the cluster.
Also, you can choose to provide a SSH user that will be used when carbonate requires connecting to another node in the cluster to perform an operation. If this is not provided, then the current user executing the command will be chosen.
Finally, you can provide HASHING_TYPE of your cluster. Default is carbon_ch, also fnv1a_ch is supported. Please note that for using fnv1a_ch hashing you need carbon 1.0.2 or newer installed (or you need to use carbon-c-relay relay instead).
usage: carbon-hosts [-h] [-c CONFIG_FILE] [-C CLUSTER]
Return the addresses for all nodes in a cluster
optional arguments:
-h, --help show this help message and exit
-c CONFIG_FILE, --config-file CONFIG_FILE
Config file to use (default:
/opt/graphite/conf/carbonate.conf)
-C CLUSTER, --cluster CLUSTER
Cluster name (default: main)
usage: carbon-lookup [-h] [-c CONFIG_FILE] [-C CLUSTER] [-s] METRIC
Lookup where a metric lives in a carbon cluster
positional arguments:
METRIC Full metric name to search for
optional arguments:
-h, --help show this help message and exit
-c CONFIG_FILE, --config-file CONFIG_FILE
Config file to use (default:
/opt/graphite/conf/carbonate.conf)
-C CLUSTER, --cluster CLUSTER
Cluster name (default: main)
-a AGGREGATION_RULES, --aggregation-rules AGGREGATION_RULES
File containing rules used in conjunction with the
"aggregated-consistent-hashing" relay method (default:
/opt/graphite/conf/aggregation-rules.conf)
-s, --short Only display the address, without port and cluster
name (default: False)
usage: carbon-list [-h] [-c CONFIG_FILE] [-C CLUSTER] [-d STORAGE_DIR]
List the metrics this carbon node contains
optional arguments:
-h, --help show this help message and exit
-c CONFIG_FILE, --config-file CONFIG_FILE
Config file to use (default:
/opt/graphite/conf/carbonate.conf)
-C CLUSTER, --cluster CLUSTER
Cluster name (default: main)
-d STORAGE_DIR, --storage-dir STORAGE_DIR
Storage dir (default: /opt/graphite/storage/whisper)
usage: carbon-sieve [-h] [-c CONFIG_FILE] [-C CLUSTER] [-f METRICS_FILE]
[-n NODE] [-I]
Given a list of metrics, output those that belong to a node
optional arguments:
-h, --help show this help message and exit
-c CONFIG_FILE, --config-file CONFIG_FILE
Config file to use (default:
/opt/graphite/conf/carbonate.conf)
-C CLUSTER, --cluster CLUSTER
Cluster name (default: main)
-a AGGREGATION_RULES, --aggregation-rules AGGREGATION_RULES
File containing rules used in conjunction with the
"aggregated-consistent-hashing" relay method (default:
/opt/graphite/conf/aggregation-rules.conf)
-f METRICS_FILE, --metrics-file METRICS_FILE
File containing metric names to filter, or '-' to read
from STDIN (default: -)
-n NODE, --node NODE Filter for metrics belonging to this node (default:
self)
-I, --invert Invert the sieve, match metrics that do NOT belong to
a node (default: False)
usage: carbon-sync [-h] [-c CONFIG_FILE] [-C CLUSTER] [-f METRICS_FILE] -s
SOURCE_NODE [-d STORAGE_DIR] [-b BATCH_SIZE]
[--source-storage-dir SOURCE_STORAGE_DIR]
[--rsync-options RSYNC_OPTIONS] [--rsync-disable-copy-dest]
[--tmpdir TMP_STAGING_DIR] [--rsync-max-retries MAX_RETRIES]
[--rsync-retries-interval SECONDS] [--dirty] [-l] [-o]
Sync local metrics using remote nodes in the cluster
optional arguments:
-h, --help show this help message and exit
-c CONFIG_FILE, --config-file CONFIG_FILE
Config file to use (env: CARBONATE_CONFIG) (default:
/opt/graphite/conf/carbonate.conf)
-C CLUSTER, --cluster CLUSTER
Cluster name (env: CARBONATE_CLUSTER) (default: main)
-f METRICS_FILE, --metrics-file METRICS_FILE
File containing metric names to filter, or '-' to read
from STDIN (default: -)
-s SOURCE_NODE, --source-node SOURCE_NODE
Override the source for metrics data (default: None)
-d STORAGE_DIR, --storage-dir STORAGE_DIR
Storage dir (default: /opt/graphite/storage/whisper)
-b BATCH_SIZE, --batch-size BATCH_SIZE
Batch size for the rsync job (default: 1000)
--source-storage-dir SOURCE_STORAGE_DIR
Source storage dir (default:
/opt/graphite/storage/whisper)
--rsync-options RSYNC_OPTIONS
Pass option(s) to rsync. Make sure to use "--rsync-
options=" if option starts with '-' (default: -azpS)
--rsync-disable-copy-dest
Avoid --copy-dest, transfer all whisper data between
nodes. (default: False)
--rsync-max-retries RETRIES
Number of times rsync will attempt to copy each batch
of metrics before moving on. If all retry attempts are
unsuccessful, carbon-sync will write a file containing
the name of each metric in the failed batch so they can
be easily retried at a later time. (Default: 3)
--rsync-retries-interval SECONDS
How long to wait in between each rsync retry attempt
(see --rsync-max-retries). (default: 5)
-t TMP_STAGING_DIR, --tmpdir TMP_STAGING_DIR
Specify an alternate location in which the temporary
rsync staging dirs will be created. This can be useful
for large syncs where the default location (as chosen
by mkdtemp) resides on a filesystem that's too small
to store all the metrics being copied from the remote
host.
--dirty If set, don't clean temporary rsync directory
(default: False)
-l, --lock Lock whisper files during filling (default: False)
-o, --overwrite Write all non nullpoints from src to dst (default:
False)
usage: carbon-path [-h] [-c CONFIG_FILE] [-C CLUSTER] [-f METRICS_FILE] [-r]
[-p] [-d STORAGE_DIR]
Transform metric paths to (or from) filesystem paths
optional arguments:
-h, --help show this help message and exit
-c CONFIG_FILE, --config-file CONFIG_FILE
Config file to use (default:
/opt/graphite/conf/carbonate.conf)
-C CLUSTER, --cluster CLUSTER
Cluster name (default: main)
-f METRICS_FILE, --metrics-file METRICS_FILE
File containing metric names to transform to file
paths, or '-' to read from STDIN (default: -)
-r, --reverse Transform from file paths to metric paths (default:
False)
-p, --prepend Prepend storage dir to file paths (default: False)
-d STORAGE_DIR, --storage-dir STORAGE_DIR
Whisper storage directory to prepend when -p given
(default: /opt/graphite/storage/whisper)
usage: carbon-stale [-h] [-c CONFIG_FILE] [-C CLUSTER] [-f METRICS_FILE] [-r]
[-d STORAGE_DIR] [-l HOURS] [-o HOURS] [-w] [-p]
Find and list potentially stale metrics.
optional arguments:
-h, --help show this help message and exit
-c CONFIG_FILE, --config-file CONFIG_FILE
Config file to use (default:
/opt/graphite/conf/carbonate.conf)
-C CLUSTER, --cluster CLUSTER
Cluster name (default: main)
-f METRICS_FILE, --metrics-file METRICS_FILE
File containing metric names to transform to file
paths, or '-' to read from STDIN (default: -)
-r, --reverse Output metrics which are not stale instead (default:
False)
-d STORAGE_DIR, --storage-dir STORAGE_DIR
Whisper storage directory to prepend when -p given
(default: /opt/graphite/storage/whisper)
-l HOURS, --limit HOURS
Definition of staleness, in hours (default: 24)
-o HOURS, --offset HOURS
Use a whisper data window ending HOURS ago (implies
-w) (default: 0)
-w, --whisper Use whisper data instead of filesystem stat() call
(default: False)
-p, --paths Print filesystem paths instead of metric names
(default: False)
usage: whisper-aggregate [-h] [-f METRICS_FILE] [-d STORAGE_DIR]
Set aggregation for whisper-backed metrics this carbon instance contains
optional arguments:
-h, --help show this help message and exit
-f METRICS_FILE, --metrics-file METRICS_FILE
File containing metric names and aggregation modes, or
'-' to read from STDIN (default: -)
-d STORAGE_DIR, --storage-dir STORAGE_DIR
Whisper storage directory (default:
/opt/graphite/storage/whisper)
usage: whisper-fill [-h] [-l] [-o] SRC DST
Backfill datapoints from one whisper file into another
positional arguments:
SRC Whisper source file
DST Whisper destination file
optional arguments:
-h, --help show this help message and exit
-l, --lock Lock whisper files during filling (default: False)
-o, --overwrite Write all non nullpoints from src to dst (default: False)
#!/bin/sh
#
# Resync a node from other nodes in the cluster
#
LOCAL_IP="$1"
for h in $(carbon-hosts) ; do
(
ssh $h -- carbon-list |
carbon-sieve -n $LOCAL_IP |
carbon-sync -s $h
) &
done
#!/bin/sh
#
# Rebalance a cluster from one size to another. Remember to cleanup metrics
# that no longer belong when all nodes are rebalanced!
#
LOCAL_IP="$1"
OLD_CLUSTER="old"
NEW_CLUSTER="main"
for h in $(carbon-hosts -C "$OLD_CLUSTER") ; do
ssh $h -- carbon-list |
carbon-sieve -C "$NEW_CLUSTER" -n $LOCAL_IP |
carbon-sync -s $h
done
#!/bin/sh
#
# List metrics from disk that don't belong
#
LOCAL_IP="$1"
carbon-list | carbon-sieve -I -n $LOCAL_IP
Metrics with whisper data that is entirely blank for the last 2 hours (perhaps useful if you suspect issues with fs timestamps or carbon clients writing in 'the future'):
carbon-list | carbon-stale --whisper --limit=2
Metrics whose metrics files appear untouched for 48 hours or more (functionally
identical to find /your/data/dir -type f -mtime +2):
carbon-list | carbon-stale --limit=48
More interesting is if you use carbon-stale, then sieve to identify stale
metrics that don't belong here (vs un-stale metrics that do belong here but
are misreported in carbon-sieve due to things like doubled-up periods in metric
paths due to broken collectors. It's a thing.)
carbon-list | carbon-stale --limit=48 | carbon-sieve -I -n $LOCAL_IP
To print file paths for use with e.g. xargs rm or whatnot, use -p:
carbon-list | carbon-stale -p | xargs -n 100 rm
The code is available under the MIT license.