This repository has no description
consul: need 3 nodes for bootstrap, needs 2 nodes for running
author
lassulus
date
(Jan 12, 2026, 2:44 PM CET)
commit
e9135750
e91357503bdcbb3c9a29c2348a5b9ce1301cf3a7
parent
fda4db14
fda4db14cfe7ee632531699b78cbd47f39ce8077
+10
-9
+10
-9
2configs/consul.nix
+10
-9
2configs/consul.nix
···
12
12
# interface.bind = "retiolum";
13
13
extraConfig = {
14
14
bind_addr = config.krebs.build.host.nets.retiolum.ip4.addr;
15
-
bootstrap_expect = 1;
15
+
bootstrap_expect = 3;
16
16
server = true;
17
17
# retry_join = config.services.consul.extraConfig.start_join;
18
18
retry_join = lib.mapAttrsToList (_n: h: lib.head h.nets.retiolum.aliases) (
···
29
29
cleanup_dead_servers = true;
30
30
# How long a server must be stable before promoting
31
31
server_stabilization_time = "10s";
32
-
# Only requires 3 servers minimum instead of all configured servers
33
-
min_quorum = 3;
32
+
# Allow cluster to operate with 2 of 3 nodes (majority quorum)
33
+
min_quorum = 2;
34
34
};
35
35
36
-
# Performance tuning for faster leader elections
36
+
# Performance tuning - slightly relaxed for network tolerance
37
37
performance = {
38
-
# Reduce raft timing for faster recovery
39
-
raft_multiplier = 1;
38
+
# 2 = default, gives more time for slow/unstable networks
39
+
# 1 was too aggressive and caused leader flapping
40
+
raft_multiplier = 2;
40
41
};
42
+
43
+
# Reconnect timeout for failed nodes (default 72h, reduce for faster cleanup)
44
+
reconnect_timeout = "24h";
41
45
42
46
# Leave on terminate for cleaner shutdowns
43
47
leave_on_terminate = true;
···
48
52
systemd.services.consul = {
49
53
# Add pre-start script to clean up potential issues
50
54
preStart = ''
51
-
# Remove any stale peers.json from failed recovery attempts
52
-
rm -f /var/lib/consul/raft/peers.json
53
-
54
55
# Ensure proper permissions
55
56
chown -R consul:consul /var/lib/consul
56
57
'';