depot/cluster/services/patroni/test.nix

150 lines
5.7 KiB
Nix

{ cluster, ... }:
let
createNode = index: { pkgs, ... }:
{
networking.firewall.allowedTCPPorts = [ 5432 8008 5010 ];
environment.systemPackages = [ pkgs.jq ];
services.patroni = {
enable = true;
softwareWatchdog = true;
settings = {
bootstrap = {
dcs = {
ttl = 30;
loop_wait = 10;
retry_timeout = 10;
maximum_lag_on_failover = 1048576;
};
initdb = [
{ encoding = "UTF8"; }
"data-checksums"
];
};
postgresql = {
use_pg_rewind = true;
use_slots = true;
authentication = {
replication = {
username = "replicator";
};
superuser = {
username = "postgres";
};
rewind = {
username = "rewind";
};
};
parameters = {
wal_level = "replica";
hot_standby_feedback = "on";
unix_socket_directories = "/tmp";
};
pg_hba = [
"host replication replicator 192.168.1.0/24 md5"
# Unsafe, do not use for anything other than tests
"host all all 0.0.0.0/0 trust"
];
};
etcd3 = {
host = "192.168.1.4:2379";
};
};
environmentFiles = {
PATRONI_REPLICATION_PASSWORD = pkgs.writeText "replication-password" "postgres";
PATRONI_SUPERUSER_PASSWORD = pkgs.writeText "superuser-password" "postgres";
PATRONI_REWIND_PASSWORD = pkgs.writeText "rewind-password" "postgres";
};
};
# We always want to restart so the tests never hang
systemd.services.patroni.serviceConfig.StartLimitIntervalSec = 0;
};
clusterName = "poseidon";
link = cluster.config.links.patroni-pg-access;
in
{
defaults = { depot, pkgs, ... }: {
environment.systemPackages = [
pkgs.jq
depot.packages.postgresql
];
services.patroni.settings.postgresql.pg_hba = [
"host all all 0.0.0.0/0 trust"
];
};
# taken from https://github.com/phfroidmont/nixpkgs/blob/patroni-module/nixos/tests/patroni.nix
testScript = ''
import json
nodeNames = json.loads('${builtins.toJSON cluster.config.services.patroni.nodes.worker}')
clientNames = json.loads('${builtins.toJSON cluster.config.services.patroni.nodes.haproxy}')
nodes = [ n for n in machines if n.name in nodeNames ]
clients = [ n for n in machines if n.name in clientNames ]
def wait_for_all_nodes_ready(expected_replicas=2):
booted_nodes = filter(lambda node: node.booted, nodes)
for node in booted_nodes:
print(node.succeed("patronictl list ${clusterName}"))
node.wait_until_succeeds(f"[ $(patronictl list -f json ${clusterName} | jq 'length') == {expected_replicas + 1} ]")
node.wait_until_succeeds("[ $(patronictl list -f json ${clusterName} | jq 'map(select(.Role | test(\"^Leader$\"))) | map(select(.State | test(\"^running$\"))) | length') == 1 ]")
node.wait_until_succeeds(f"[ $(patronictl list -f json ${clusterName} | jq 'map(select(.Role | test(\"^Replica$\"))) | map(select(.State | test(\"^streaming$\"))) | length') == {expected_replicas} ]")
print(node.succeed("patronictl list ${clusterName}"))
for client in clients:
client.wait_until_succeeds("psql -h ${link.ipv4} -p ${link.portStr} -U postgres --command='select 1;'")
def run_dummy_queries():
for client in clients:
client.succeed("psql -h ${link.ipv4} -p ${link.portStr} -U postgres --pset='pager=off' --tuples-only --command='insert into dummy(val) values (101);'")
client.succeed("test $(psql -h ${link.ipv4} -p ${link.portStr} -U postgres --pset='pager=off' --tuples-only --command='select val from dummy where val = 101;') -eq 101")
client.succeed("psql -h ${link.ipv4} -p ${link.portStr} -U postgres --pset='pager=off' --tuples-only --command='delete from dummy where val = 101;'")
start_all()
with subtest("should bootstrap a new patroni cluster"):
wait_for_all_nodes_ready()
with subtest("should be able to insert and select"):
clients[0].succeed("psql -h ${link.ipv4} -p ${link.portStr} -U postgres --command='create table dummy as select * from generate_series(1, 100) as val;'")
for client in clients:
client.succeed("test $(psql -h ${link.ipv4} -p ${link.portStr} -U postgres --pset='pager=off' --tuples-only --command='select count(distinct val) from dummy;') -eq 100")
with subtest("should restart after all nodes are crashed"):
for node in nodes:
node.crash()
for node in nodes:
node.start()
wait_for_all_nodes_ready()
with subtest("should be able to run queries while any one node is crashed"):
masterNodeName = nodes[0].succeed("patronictl list -f json ${clusterName} | jq '.[] | select(.Role | test(\"^Leader$\")) | .Member' -r").strip()
masterNodeIndex = next((i for i, v in enumerate(nodes) if v.name == masterNodeName), None)
# Move master node at the end of the list to avoid multiple failovers (makes the test faster and more consistent)
nodes.append(nodes.pop(masterNodeIndex))
for node in nodes:
node.crash()
wait_for_all_nodes_ready(1)
# Execute some queries while a node is down.
run_dummy_queries()
# Restart crashed node.
node.start()
wait_for_all_nodes_ready()
# Execute some queries with the node back up.
run_dummy_queries()
'';
}