nixpkgs-suyu/nixos/tests/patroni.nix
Paul-Henri Froidmont d2ce318bd4 nixos/patroni: init
2022-08-16 18:57:15 +02:00

204 lines
7 KiB
Nix

import ./make-test-python.nix ({ pkgs, lib, ... }:
let
nodesIps = [
"192.168.1.1"
"192.168.1.2"
"192.168.1.3"
];
createNode = index: { pkgs, ... }:
let
ip = builtins.elemAt nodesIps index; # since we already use IPs to identify servers
in
{
networking.interfaces.eth1.ipv4.addresses = pkgs.lib.mkOverride 0 [
{ address = ip; prefixLength = 16; }
];
networking.firewall.allowedTCPPorts = [ 5432 8008 5010 ];
environment.systemPackages = [ pkgs.jq ];
services.patroni = {
enable = true;
postgresqlPackage = pkgs.postgresql_14.withPackages (p: [ p.pg_safeupdate ]);
scope = "cluster1";
name = "node${toString(index + 1)}";
nodeIp = ip;
otherNodesIps = builtins.filter (h: h != ip) nodesIps;
softwareWatchdog = true;
settings = {
bootstrap = {
dcs = {
ttl = 30;
loop_wait = 10;
retry_timeout = 10;
maximum_lag_on_failover = 1048576;
};
initdb = [
{ encoding = "UTF8"; }
"data-checksums"
];
};
postgresql = {
use_pg_rewind = true;
use_slots = true;
authentication = {
replication = {
username = "replicator";
};
superuser = {
username = "postgres";
};
rewind = {
username = "rewind";
};
};
parameters = {
listen_addresses = "${ip}";
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;
};
in
{
name = "patroni";
nodes = {
node1 = createNode 0;
node2 = createNode 1;
node3 = createNode 2;
etcd = { pkgs, ... }: {
networking.interfaces.eth1.ipv4.addresses = pkgs.lib.mkOverride 0 [
{ address = "192.168.1.4"; prefixLength = 16; }
];
services.etcd = {
enable = true;
listenClientUrls = [ "http://192.168.1.4:2379" ];
};
networking.firewall.allowedTCPPorts = [ 2379 ];
};
client = { pkgs, ... }: {
environment.systemPackages = [ pkgs.postgresql_14 ];
networking.interfaces.eth1.ipv4.addresses = pkgs.lib.mkOverride 0 [
{ address = "192.168.2.1"; prefixLength = 16; }
];
services.haproxy = {
enable = true;
config = ''
global
maxconn 100
defaults
log global
mode tcp
retries 2
timeout client 30m
timeout connect 4s
timeout server 30m
timeout check 5s
listen cluster1
bind 127.0.0.1:5432
option httpchk
http-check expect status 200
default-server inter 3s fall 3 rise 2 on-marked-down shutdown-sessions
${builtins.concatStringsSep "\n" (map (ip: "server postgresql_${ip}_5432 ${ip}:5432 maxconn 100 check port 8008") nodesIps)}
'';
};
};
};
testScript = ''
nodes = [node1, node2, node3]
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 cluster1"))
node.wait_until_succeeds(f"[ $(patronictl list -f json cluster1 | jq 'length') == {expected_replicas + 1} ]")
node.wait_until_succeeds("[ $(patronictl list -f json cluster1 | jq 'map(select(.Role | test(\"^Leader$\"))) | map(select(.State | test(\"^running$\"))) | length') == 1 ]")
node.wait_until_succeeds(f"[ $(patronictl list -f json cluster1 | jq 'map(select(.Role | test(\"^Replica$\"))) | map(select(.State | test(\"^running$\"))) | length') == {expected_replicas} ]")
print(node.succeed("patronictl list cluster1"))
client.wait_until_succeeds("psql -h 127.0.0.1 -U postgres --command='select 1;'")
def run_dummy_queries():
client.succeed("psql -h 127.0.0.1 -U postgres --pset='pager=off' --tuples-only --command='insert into dummy(val) values (101);'")
client.succeed("test $(psql -h 127.0.0.1 -U postgres --pset='pager=off' --tuples-only --command='select val from dummy where val = 101;') -eq 101")
client.succeed("psql -h 127.0.0.1 -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"):
client.succeed("psql -h 127.0.0.1 -U postgres --command='create table dummy as select * from generate_series(1, 100) as val;'")
client.succeed("test $(psql -h 127.0.0.1 -U postgres --pset='pager=off' --tuples-only --command='select count(distinct val) from dummy;') -eq 100")
with subtest("should restart after all nodes are crashed"):
for node in nodes:
node.crash()
for node in nodes:
node.start()
wait_for_all_nodes_ready()
with subtest("should be able to run queries while any one node is crashed"):
masterNodeName = node1.succeed("patronictl list -f json cluster1 | jq '.[] | select(.Role | test(\"^Leader$\")) | .Member' -r").strip()
masterNodeIndex = int(masterNodeName[len(masterNodeName)-1]) - 1
# 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()
'';
})