centralized kubequery information in Postgres database.
Create a centralized 'loader' that polls a collection of kubernetes clusters using kubequery and store all results in a single Postgres database using the same schema as kubequery.
Schema: https://github.com/Uptycs/kubequery/blob/master/docs/schema.md
Kubequery is usually installed on specific clusters and used to make 'live' queries against the local kubernetes resources of that cluster using SQL. This is a different approach where this software and kubequery is installed on only one cluster(the aggregator) and polls other clusters(the targets) using kubequery remote-access. The results are saved in a centralized Postgres database.
Benefits:
-
Store kube resources from hundreds of clusters in a single Postgres database.
-
Almost real-time queries against hundreds of clusters since all the data is in the same Postgres database.
-
Index can be added to the PG database for improved access time.
-
Polling the clusters and saving the results in the database is fast and can be done every few minutes. Clusters can be polled in parallel for improved speed.
-
On large clusters, SQL joins can be really slow against the sqlite database used by kubequery. The Postgres tables are not virtual and can be optimized to solve this problem.
-
Columns in the kubequery schema like labels and annotations are TEXT but contain json information. These columns can be converted to Postgres JSONB for easy access.
-
Only the basic "select * from kubernetes_.." queries are executed by kubequery on the target clusters during aggregation. All complex queries (JOINS, multi-cluster) are done on the Postgres database.
-
Note: For speed, the Postgres database should be located on the aggregator cluster.
For this application, kubequery had to be modified to support remote cluster access. Currently, kubequery assumes that it is running inside the k8s cluster being probed. This change allows kubequery to connect to a remote cluster and extract resource data. See ./kubequery/ for details.
Kubequery script using parameters generated by the loader:
#!/bin/sh
# runquery <table> <token> <cluster_addr> <cluster> <uuid> <ix>
TABLE=$1
export KUBEQUERY_TOKEN=$2
export KUBEQUERY_ADDR=$3
export CLUSTER_NAME=$4
export CLUSTER_UID=$5
export CLUSTER_IX=$6
echo "select * from $TABLE;" | /opt/uptycs/bin/basequery --flagfile=/opt/uptycs/etc/kubequery.flags --config_path=/opt/uptycs/etc/kubequery.conf --extensions_socket=/opt/uptycs/var/kubequeryi.em$TABLE.$CLUSTER_IX --extensions_autoload=/opt/uptycs/etc/autoload.exts --extensions_require=kubequery --extension_event_tables=kubernetes_events --disable_database --json --disable_events=false -S > /tmp/${CLUSTER_NAME}-$TABLE.json
# Count namespaces and pods by Cluster
=> select cluster_name , count(distinct namespace) ns_count , count(*) pod_cnt from kubernetes_pods group by cluster_name;
cluster_name | ns_count | pod_cnt
-----------------------+---------+---------
national1 | 343 | 9636
national2 | 43 | 30324
national3 | 29 | 9988
regional1 | 49 | 8342
regional2 | 50 | 6845
regional4 | 16 | 2943
regional5 | 33 | 3062
regional6 | 38 | 4489
(11 rows)
Time: 314.390 ms
# MOST POPULAR CONTAINER IMAGES
=> select c.image, count(*) pod_cnt from kubernetes_pod_containers c group by c.image order by count(*) desc limit 10;
image | pod_cnt
--------------------------------------------------------------+--------
hub.obfusca.net/xxxxx-packager/system/system:3.5.1-1 | 18080
hub.obfusca.net/xxxxx-arch/varnish-image:7.0.2-dev1 | 11376
hub.obfusca.net/xxxxx-packager/system/system:3.3.2-1 | 10820
hub.obfusca.net/library/telegraf:1.13.4 | 5434
registry.xxxxxxx.net/anchorfree/twemproxy:latest | 5384
hub.obfusca.net/rio/services/rccs-decoder:1.1.13_959 | 4217
hub.obfusca.net/k8s-eng/xxxxx/rdei/k8s-dns-node-cache:1.21.1 | 3727
hub.obfusca.net/xxxxx/rdei/node-exporter:v1.3.1 | 3382
hub.obfusca.net/k8s-eng/xxxxx/rdei/sumatra:0.42.07 | 3315
hub.obfusca.net/rio/services/bmw:1.9.0_1473 | 3299
(10 rows)
Time: 230.233 ms
- Add the postgres schema from database/ to your Postgres instance.
- Create the custom image (uptycs/kubequery:1.1.1-remote) using the info from kubequery/
- Create the kubernetes Deployment using the info from kubernetes/
# concurrency=5
$ /loader -c 5
4 Cluster START testcluster-01 - https://testcluster-01:6443
1 Cluster START testcluster2-01 - https://testcluster2-01:6443
2 Cluster START testcluster3-01 - https://testcluster3-01:6443
3 Cluster START testcluster5-01 - https://testcluster5-01:6443
0 Cluster START testcluster6-01 - https://testcluster-02:6443
------------------------------------------------------------------------------
2 Cluster REPORT testcluster3-01 - https://testcluster3-01:6443
------------------------------------------------------------------------------
testcluster3-01 kubernetes_namespaces TABLE
testcluster3-01 kubernetes_namespaces Rows=19, Duration=1.05929269s
testcluster3-01 kubernetes_nodes TABLE
testcluster3-01 kubernetes_nodes Rows=6, Duration=976.341347ms
testcluster3-01 kubernetes_pods TABLE
testcluster3-01 kubernetes_pods Rows=204, Duration=1.688286795s
testcluster3-01 kubernetes_pod_containers TABLE
testcluster3-01 kubernetes_pod_containers Rows=274, Duration=1.883914113s
5 Cluster START testcluster5 - https://testcluster5:6443
------------------------------------------------------------------------------
4 Cluster REPORT testcluster-01 - https://testcluster-01:6443
------------------------------------------------------------------------------
testcluster-01 kubernetes_namespaces TABLE
testcluster-01 kubernetes_namespaces Rows=33, Duration=894.902131ms
testcluster-01 kubernetes_nodes TABLE
testcluster-01 kubernetes_nodes Rows=60, Duration=970.209593ms
testcluster-01 kubernetes_pods TABLE
testcluster-01 kubernetes_pods Rows=647, Duration=2.524690614s
testcluster-01 kubernetes_pod_containers TABLE
testcluster-01 kubernetes_pod_containers Rows=953, Duration=3.389727493s
...
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent