Refactor infra
Docker / build-and-push-image (backend, backend, coracle/caravel-backend) (push) Failing after 5m6s
Docker / build-and-push-image (frontend, frontend, coracle/caravel-frontend) (push) Successful in 2m44s

This commit is contained in:
Jon Staab
2026-05-15 13:52:35 -07:00
parent cfa52d739f
commit 7134915665
+157 -255
View File
@@ -54,15 +54,17 @@ impl Infra {
}
async fn handle_activity(&self, activity: &Activity) -> Result<()> {
let needs_sync = should_sync_relay_activity(activity.activity_type.as_str());
let needs_sync = matches!(
activity.activity_type.as_str(),
"create_relay" | "update_relay" | "activate_relay" | "deactivate_relay" | "fail_relay_sync"
);
if !needs_sync || activity.resource_type != "relay" {
if activity.resource_type != "relay" || !needs_sync {
return Ok(());
}
if activity.activity_type == "fail_relay_sync" {
self.schedule_relay_sync_retry(&activity.resource_id, "activity")
.await?;
self.schedule_relay_sync_retry(&activity.resource_id, "activity").await?;
return Ok(());
}
@@ -70,9 +72,7 @@ impl Infra {
return Ok(());
};
let is_new = self.relay_sync_is_new(&relay).await?;
self.sync_and_report(&relay, is_new).await;
self.sync_relay(&relay).await;
Ok(())
}
@@ -87,8 +87,7 @@ impl Infra {
for relay in relays {
if relay.sync_error.trim().is_empty() {
let is_new = self.relay_sync_is_new(&relay).await?;
self.sync_and_report(&relay, is_new).await;
self.sync_relay(&relay).await;
} else {
self.schedule_relay_sync_retry(&relay.id, source).await?;
}
@@ -98,10 +97,28 @@ impl Infra {
}
async fn schedule_relay_sync_retry(&self, relay_id: &str, source: &str) -> Result<()> {
let activities = self.query.list_activity_for_relay(relay_id).await?;
let consecutive_failures = consecutive_sync_failures(&activities);
fn get_retry_delay(consecutive_failures: usize) -> Option<Duration> {
let retry_attempt = consecutive_failures.max(1);
if retry_attempt > RELAY_SYNC_RETRY_MAX_ATTEMPTS {
return None;
}
let Some(delay) = relay_sync_retry_delay(consecutive_failures) else {
let exponent = (retry_attempt - 1).min(31);
let multiplier = 1u64 << exponent;
let delay_secs = RELAY_SYNC_RETRY_BASE_DELAY_SECS
.saturating_mul(multiplier)
.min(RELAY_SYNC_RETRY_MAX_DELAY_SECS);
Some(Duration::from_secs(delay_secs))
}
let activities = self.query.list_activity_for_relay(relay_id).await?;
let consecutive_failures = activities
.iter()
.take_while(|activity| activity.activity_type == "fail_relay_sync")
.count();
let Some(delay) = get_retry_delay(consecutive_failures) else {
tracing::warn!(
relay = relay_id,
consecutive_failures,
@@ -125,40 +142,20 @@ impl Infra {
tokio::spawn(async move {
tokio::time::sleep(delay).await;
if let Err(e) = infra.retry_relay_sync(&relay_id).await {
tracing::error!(relay = %relay_id, error = %e, "relay sync retry task failed");
match infra.query.get_relay(&relay_id).await {
Ok(Some(relay)) => infra.sync_relay(&relay).await,
Ok(None) => {}
Err(e) => {
tracing::error!(relay = %relay_id, error = %e, "relay sync retry task failed");
}
}
});
Ok(())
}
async fn retry_relay_sync(&self, relay_id: &str) -> Result<()> {
let Some(relay) = self.query.get_relay(relay_id).await? else {
return Ok(());
};
if relay.sync_error.trim().is_empty() {
tracing::debug!(relay = %relay.id, "skip relay sync retry; relay has no sync_error");
return Ok(());
}
let is_new = self.relay_sync_is_new(&relay).await?;
self.sync_and_report(&relay, is_new).await;
Ok(())
}
async fn relay_sync_is_new(&self, relay: &Relay) -> Result<bool> {
if relay.synced == 1 {
return Ok(false);
}
let has_completed_sync = self.query.relay_has_completed_sync(&relay.id).await?;
Ok(!has_completed_sync)
}
async fn sync_and_report(&self, relay: &Relay, is_new: bool) {
match self.sync_relay(relay, is_new).await {
async fn sync_relay(&self, relay: &Relay) {
match self.try_sync_relay(relay).await {
Ok(()) => {
tracing::info!(relay = %relay.id, "relay sync succeeded");
if let Err(e) = self.command.complete_relay_sync(&relay.id).await {
@@ -174,229 +171,134 @@ impl Infra {
}
}
pub async fn list_relay_members(&self, relay_id: &str) -> Result<Vec<String>> {
let client = reqwest::Client::new();
let base = self.env.zooid_api_url.trim_end_matches('/');
let url = format!("{base}/relay/{relay_id}/members");
let auth = self.env.make_auth(&url, HttpMethod::GET).await?;
async fn try_sync_relay(&self, relay: &Relay) -> Result<()> {
// A relay is "new" (POST with a freshly generated secret) only if it has
// never completed a sync. `synced == 1` short-circuits the activity lookup;
// otherwise check the activity history so that a re-sync after an update
// (which resets `synced` to 0) PATCHes instead of clobbering the secret.
let is_new = relay.synced != 1
&& !self.query.relay_has_completed_sync(&relay.id).await?;
let response = client
.get(&url)
.header("Authorization", auth)
.send()
.await?;
let mut body = serde_json::json!({
"host": format!("{}.{}", relay.subdomain, self.env.relay_domain),
"schema": relay.schema,
"inactive": relay.status == RELAY_STATUS_INACTIVE
|| relay.status == RELAY_STATUS_DELINQUENT,
"info": {
"name": relay.info_name,
"icon": relay.info_icon,
"description": relay.info_description,
"pubkey": relay.tenant,
},
"policy": {
"public_join": relay.policy_public_join == 1,
"strip_signatures": relay.policy_strip_signatures == 1,
},
"groups": { "enabled": relay.groups_enabled == 1 },
"management": { "enabled": relay.management_enabled == 1 },
"blossom": if relay.blossom_enabled == 1 {
serde_json::json!({
"enabled": true,
"adapter": "s3",
"s3": {
"endpoint": self.env.blossom_s3_endpoint,
"region": self.env.blossom_s3_region,
"bucket": self.env.blossom_s3_bucket,
"access_key": self.env.blossom_s3_access_key,
"secret_key": self.env.blossom_s3_secret_key,
"key_prefix": relay.schema,
},
})
} else {
serde_json::json!({ "enabled": false })
},
"livekit": if relay.livekit_enabled == 1 {
serde_json::json!({
"enabled": true,
"server_url": self.env.livekit_url,
"api_key": self.env.livekit_api_key,
"api_secret": self.env.livekit_api_secret,
})
} else {
serde_json::json!({ "enabled": false })
},
"push": { "enabled": relay.push_enabled == 1 },
"roles": {
"admin": { "can_manage": true, "can_invite": true },
"member": { "can_invite": true },
},
});
if !response.status().is_success() {
let status = response.status();
let body = response.text().await.unwrap_or_default();
anyhow::bail!("zooid members returned {status}: {body}");
// Only provide a secret if the relay is new. This allows us to not store the relay secrets on our side.
if is_new {
if let Some(obj) = body.as_object_mut() {
obj.insert(
"secret".to_string(),
serde_json::Value::String(Keys::generate().secret_key().to_secret_hex()),
);
}
}
let body = response.text().await?;
parse_relay_members_response(&body)
}
async fn sync_relay(&self, relay: &Relay, is_new: bool) -> Result<()> {
let client = reqwest::Client::new();
let base = self.env.zooid_api_url.trim_end_matches('/');
let host = if self.env.relay_domain.is_empty() {
relay.subdomain.clone()
} else {
format!("{}.{}", relay.subdomain, self.env.relay_domain)
};
let livekit = if relay.livekit_enabled == 1 {
serde_json::json!({
"enabled": true,
"server_url": self.env.livekit_url,
"api_key": self.env.livekit_api_key,
"api_secret": self.env.livekit_api_secret,
})
} else {
serde_json::json!({ "enabled": false })
};
let body = relay_sync_body(
relay,
host,
livekit,
is_new.then(|| Keys::generate().secret_key().to_secret_hex()),
&self.env,
);
let url = format!("{}/relay/{}", base, relay.id);
let auth = self
.env
.make_auth(&url, zooid_sync_http_method(is_new))
let method = if is_new { HttpMethod::POST } else { HttpMethod::PATCH };
self.request(method, &format!("relay/{}", relay.id), Some(&body))
.await?;
let request = if is_new {
client.post(&url)
} else {
client.patch(&url)
};
let response = request
.header("Authorization", auth)
.json(&body)
.send()
.await?;
if !response.status().is_success() {
let status = response.status();
let body = response.text().await.unwrap_or_default();
anyhow::bail!("zooid sync returned {status}: {body}")
}
Ok(())
}
}
fn zooid_sync_http_method(is_new: bool) -> HttpMethod {
if is_new {
HttpMethod::POST
} else {
HttpMethod::PATCH
pub async fn list_relay_members(&self, relay_id: &str) -> Result<Vec<String>> {
#[derive(serde::Deserialize)]
struct MembersResponse {
members: Vec<String>,
}
let response = self
.request(HttpMethod::GET, &format!("relay/{relay_id}/members"), None)
.await?;
let parsed: MembersResponse = response.json().await?;
Ok(parsed.members)
}
// Internal utilities
/// Sends an authenticated request to the zooid API at `path` (relative to
/// `env.zooid_api_url`). Returns the response on 2xx; bails with the body
/// text otherwise.
async fn request(
&self,
method: HttpMethod,
path: &str,
body: Option<&serde_json::Value>,
) -> Result<reqwest::Response> {
let client = reqwest::Client::builder()
.timeout(Duration::from_secs(5))
.build()?;
let base = self.env.zooid_api_url.trim_end_matches('/');
let path = path.trim_start_matches('/');
let url = format!("{base}/{path}");
let auth = self.env.make_auth(&url, method).await?;
let reqwest_method = match method {
HttpMethod::GET => reqwest::Method::GET,
HttpMethod::POST => reqwest::Method::POST,
HttpMethod::PUT => reqwest::Method::PUT,
HttpMethod::PATCH => reqwest::Method::PATCH,
};
let mut req = client
.request(reqwest_method, &url)
.header("Authorization", auth);
if let Some(body) = body {
req = req.json(body);
}
let response = req.send().await?;
if !response.status().is_success() {
let status = response.status();
let text = response.text().await.unwrap_or_default();
anyhow::bail!("zooid {method} {path} returned {status}: {text}");
}
Ok(response)
}
}
fn parse_relay_members_response(body: &str) -> Result<Vec<String>> {
let value: serde_json::Value = serde_json::from_str(body)?;
if let Some(members) = members_from_value(&value) {
return Ok(members);
}
if let Some(members) = value.get("members").and_then(members_from_value) {
return Ok(members);
}
if let Some(members) = value
.get("data")
.and_then(|data| data.get("members"))
.and_then(members_from_value)
{
return Ok(members);
}
anyhow::bail!("zooid members response missing members array")
}
fn members_from_value(value: &serde_json::Value) -> Option<Vec<String>> {
let values = value.as_array()?;
values
.iter()
.map(|value| value.as_str().map(ToString::to_string))
.collect()
}
fn relay_sync_body(
relay: &Relay,
host: String,
livekit: serde_json::Value,
secret: Option<String>,
env: &Env,
) -> serde_json::Value {
let blossom = blossom_sync_json(relay, env);
let mut body = serde_json::json!({
"host": host,
"schema": relay.schema,
"inactive": relay.status == RELAY_STATUS_INACTIVE
|| relay.status == RELAY_STATUS_DELINQUENT,
"info": {
"name": relay.info_name,
"icon": relay.info_icon,
"description": relay.info_description,
"pubkey": relay.tenant,
},
"policy": {
"public_join": relay.policy_public_join == 1,
"strip_signatures": relay.policy_strip_signatures == 1,
},
"groups": { "enabled": relay.groups_enabled == 1 },
"management": { "enabled": relay.management_enabled == 1 },
"blossom": blossom,
"livekit": livekit,
"push": { "enabled": relay.push_enabled == 1 },
"roles": {
"admin": { "can_manage": true, "can_invite": true },
"member": { "can_invite": true },
},
});
if let (Some(secret), Some(body_obj)) = (secret, body.as_object_mut()) {
body_obj.insert("secret".to_string(), serde_json::Value::String(secret));
}
body
}
/// Relay sync sets `key_prefix` to the relay schema so each relay gets its own
/// blob namespace within the shared bucket.
fn blossom_sync_json(relay: &Relay, env: &Env) -> serde_json::Value {
if relay.blossom_enabled != 1 {
return serde_json::json!({ "enabled": false });
}
let mut s3_obj = serde_json::Map::new();
if !env.blossom_s3_endpoint.trim().is_empty() {
s3_obj.insert(
"endpoint".to_string(),
serde_json::Value::String(env.blossom_s3_endpoint.clone()),
);
}
s3_obj.insert(
"region".to_string(),
serde_json::Value::String(env.blossom_s3_region.clone()),
);
s3_obj.insert(
"bucket".to_string(),
serde_json::Value::String(env.blossom_s3_bucket.clone()),
);
s3_obj.insert(
"access_key".to_string(),
serde_json::Value::String(env.blossom_s3_access_key.clone()),
);
s3_obj.insert(
"secret_key".to_string(),
serde_json::Value::String(env.blossom_s3_secret_key.clone()),
);
s3_obj.insert(
"key_prefix".to_string(),
serde_json::Value::String(relay.schema.clone()),
);
serde_json::json!({
"enabled": true,
"adapter": "s3",
"s3": serde_json::Value::Object(s3_obj),
})
}
fn should_sync_relay_activity(activity_type: &str) -> bool {
matches!(
activity_type,
"create_relay" | "update_relay" | "activate_relay" | "deactivate_relay" | "fail_relay_sync"
)
}
fn consecutive_sync_failures(activities: &[Activity]) -> usize {
activities
.iter()
.take_while(|activity| activity.activity_type == "fail_relay_sync")
.count()
}
fn relay_sync_retry_delay(consecutive_failures: usize) -> Option<Duration> {
let retry_attempt = consecutive_failures.max(1);
if retry_attempt > RELAY_SYNC_RETRY_MAX_ATTEMPTS {
return None;
}
let exponent = (retry_attempt - 1).min(31);
let multiplier = 1u64 << exponent;
let delay_secs = RELAY_SYNC_RETRY_BASE_DELAY_SECS
.saturating_mul(multiplier)
.min(RELAY_SYNC_RETRY_MAX_DELAY_SECS);
Some(Duration::from_secs(delay_secs))
}