vinzenz/hyperhive
vinzenz/hyperhive
Watch
0
0
Fork
You've already forked hyperhive
 0
Code Activity

raise the cert role token ttl to 1h

Browse source
This commit is contained in:
atlas 2026-10-28 20:18:18 +01:00 • committed by mara
commit 099af86cd2
4 changed files with 23 additions and 13 deletions
Download patch file Download diff file Expand all files Collapse all files

12
docs/swarm/credentials.md
View file

@ -456,8 +456,8 @@ responders log in on every check, and the store keeps no record of a batch token
so unauthenticated requests don't grow its token storage. Nothing can renew so unauthenticated requests don't grow its token storage. Nothing can renew
or revoke a batch token. or revoke a batch token.
The other roles issue service tokens that live 15 minutes (`token_ttl=15m` and The other roles issue service tokens that live an hour (`token_ttl=1h` and
`token_max_ttl=15m`). That covers the roles `swarm-bao.nix` writes, and the `token_max_ttl=1h`). That covers the roles `swarm-bao.nix` writes, and the
agent and hive roles the controller writes through agent and hive roles the controller writes through
`SecretStore::write_cert_role`, which takes the same value from `SecretStore::write_cert_role`, which takes the same value from
`CERT_ROLE_TOKEN_TTL`. The store keeps a lease for every token until it `CERT_ROLE_TOKEN_TTL`. The store keeps a lease for every token until it
@ -465,10 +465,14 @@ expires, and the mount default of 32 days leaves one lease per login for a
month; at the 17 thousand logins a day this swarm made, that came to half a month; at the 17 thousand logins a day this swarm made, that came to half a
million leases. A reader logs in per operation and holds the token for seconds. million leases. A reader logs in per operation and holds the token for seconds.
The longest hold is the controller's forge work, which stops after two minutes. The longest hold is the controller's forge work, which stops after two minutes.
A client that keeps a `SecretStore` handle longer than 15 minutes must log in A client that keeps a `SecretStore` handle longer than an hour must log in
again, because nothing renews the token. A role the controller writes takes the again, because nothing renews the token. A role the controller writes takes the
TTL at its next write: a hive's at the controller's start, an agent's at its TTL at its next write: a hive's at the controller's start, an agent's at its
next certificate renewal. next certificate renewal. At that rate an hour keeps about 700 tokens live.
The granter's role keeps 15 minutes. Only the one-time bootstrap token writes
it, and a deployed store never rewrites it, so the longer TTL would not reach
it; the granter logs in and finishes within seconds.
### What a missing leaf costs ### What a missing leaf costs

2
nix/host-modules/swarm-bao.nix
View file

@ -995,7 +995,7 @@ let
# within two minutes (the controller's forge work is cut off at # within two minutes (the controller's forge work is cut off at
# `FORGE_DEADLINE`). Rust-written roles take `CERT_ROLE_TOKEN_TTL`, which # `FORGE_DEADLINE`). Rust-written roles take `CERT_ROLE_TOKEN_TTL`, which
# is the same value. # is the same value.
certRoleTokenTtl = "15m"; certRoleTokenTtl = "1h";
# The agent CA's certificate, cached on this host by `swarm-bao-agent-pki`, # The agent CA's certificate, cached on this host by `swarm-bao-agent-pki`,
# so `swarm-bao-certs` can compose the listener's bundle before the store # so `swarm-bao-certs` can compose the listener's bundle before the store

16
nix/module-eval/bao-grants.nix
View file

@ -907,8 +907,8 @@ let
&& batch "swarm-bao-metrics-auth-policy" "swarm-metrics-auth" && batch "swarm-bao-metrics-auth-policy" "swarm-metrics-auth"
&& lib.hasInfix "token_policies=swarm-ci-runner" ci && lib.hasInfix "token_policies=swarm-ci-runner" ci
&& !(lib.hasInfix "token_type" ci) && !(lib.hasInfix "token_type" ci)
&& lib.hasInfix "token_ttl=15m" ci && lib.hasInfix "token_ttl=1h" ci
&& lib.hasInfix "token_max_ttl=15m" ci; && lib.hasInfix "token_max_ttl=1h" ci;
} }
{ {
name = "the PKI unit signs the metrics store responder's leaf under its own subject"; name = "the PKI unit signs the metrics store responder's leaf under its own subject";
@ -1476,8 +1476,9 @@ let
# A cert login leaves a lease until its token expires, and a role without # A cert login leaves a lease until its token expires, and a role without
# `token_ttl` gets the mount's default of 32 days. Every role write in # `token_ttl` gets the mount's default of 32 days. Every role write in
# every unit is read, so the TTL must be on the command itself, not # every unit is read, so the TTL must be on the command itself, not
# somewhere else in the script. A role is a service token at `15m`, or a # somewhere else in the script. A role is a service token at `1h`, or a
# batch token at `5m`. # batch token at `5m`. The granter keeps `15m`: only the bootstrap token
# writes its role, so a deployed store does not rewrite it.
# #
# The presence control: the roles found include the controller's, the # The presence control: the roles found include the controller's, the
# granter's, a batch responder's, the CI runner's and a per-hive leaf # granter's, a batch responder's, the CI runner's and a per-hive leaf
@ -1508,7 +1509,12 @@ let
writes = roleWrites baoGrantHere ++ roleWrites baoGrantTwoHives; writes = roleWrites baoGrantHere ++ roleWrites baoGrantTwoHives;
short = short =
w: w:
(lib.hasInfix "token_ttl=15m" w.command && lib.hasInfix "token_max_ttl=15m" w.command) (lib.hasInfix "token_ttl=1h" w.command && lib.hasInfix "token_max_ttl=1h" w.command)
|| (
w.role == "bao-granter"
&& lib.hasInfix "token_ttl=15m" w.command
&& lib.hasInfix "token_max_ttl=15m" w.command
)
|| ( || (
lib.hasInfix "token_type=batch" w.command lib.hasInfix "token_type=batch" w.command
&& lib.hasInfix "token_ttl=5m" w.command && lib.hasInfix "token_ttl=5m" w.command

6
swarm-secret-client/src/client.rs
View file

@ -35,7 +35,7 @@ pub const DEFAULT_CERT_MOUNT: &str = "cert";
/// it leaves one lease per login for the mount's default of 32 days. /// it leaves one lease per login for the mount's default of 32 days.
/// [`SecretStore`] holds the token it logs in with and renews nothing: log in /// [`SecretStore`] holds the token it logs in with and renews nothing: log in
/// again rather than keep a handle past this. /// again rather than keep a handle past this.
pub const CERT_ROLE_TOKEN_TTL: &str = "15m"; pub const CERT_ROLE_TOKEN_TTL: &str = "1h";
/// Upper bound on one request to the store, login included. /// Upper bound on one request to the store, login included.
/// ///
@ -738,8 +738,8 @@ mod tests {
#[test] #[test]
fn a_cert_role_is_written_with_short_lived_tokens() { fn a_cert_role_is_written_with_short_lived_tokens() {
let r = cert_role_request("cert", "hive-agent-a", "PEM", "hive-agent-a", &["shared"]); let r = cert_role_request("cert", "hive-agent-a", "PEM", "hive-agent-a", &["shared"]);
assert_eq!(r.token_ttl.as_deref(), Some("15m")); assert_eq!(r.token_ttl.as_deref(), Some("1h"));
assert_eq!(r.token_max_ttl.as_deref(), Some("15m")); assert_eq!(r.token_max_ttl.as_deref(), Some("1h"));
// The control: the rest of the role is still written. // The control: the rest of the role is still written.
assert_eq!( assert_eq!(
r.allowed_common_names, r.allowed_common_names,