Line data Source code
1 : use std::collections::HashMap;
2 :
3 : use futures::Future;
4 : use pageserver_api::{
5 : controller_api::NodeRegisterRequest,
6 : shard::TenantShardId,
7 : upcall_api::{
8 : ReAttachRequest, ReAttachResponse, ReAttachResponseTenant, ValidateRequest,
9 : ValidateRequestTenant, ValidateResponse,
10 : },
11 : };
12 : use serde::{de::DeserializeOwned, Serialize};
13 : use tokio_util::sync::CancellationToken;
14 : use url::Url;
15 : use utils::{backoff, failpoint_support, generation::Generation, id::NodeId};
16 :
17 : use crate::{config::PageServerConf, virtual_file::on_fatal_io_error};
18 : use pageserver_api::config::NodeMetadata;
19 :
20 : /// The Pageserver's client for using the control plane API: this is a small subset
21 : /// of the overall control plane API, for dealing with generations (see docs/rfcs/025-generation-numbers.md)
22 : pub struct ControlPlaneClient {
23 : http_client: reqwest::Client,
24 : base_url: Url,
25 : node_id: NodeId,
26 : cancel: CancellationToken,
27 : }
28 :
29 : /// Represent operations which internally retry on all errors other than
30 : /// cancellation token firing: the only way they can fail is ShuttingDown.
31 : pub enum RetryForeverError {
32 : ShuttingDown,
33 : }
34 :
35 : pub trait ControlPlaneGenerationsApi {
36 : fn re_attach(
37 : &self,
38 : conf: &PageServerConf,
39 : ) -> impl Future<
40 : Output = Result<HashMap<TenantShardId, ReAttachResponseTenant>, RetryForeverError>,
41 : > + Send;
42 : fn validate(
43 : &self,
44 : tenants: Vec<(TenantShardId, Generation)>,
45 : ) -> impl Future<Output = Result<HashMap<TenantShardId, bool>, RetryForeverError>> + Send;
46 : }
47 :
48 : impl ControlPlaneClient {
49 : /// A None return value indicates that the input `conf` object does not have control
50 : /// plane API enabled.
51 0 : pub fn new(conf: &'static PageServerConf, cancel: &CancellationToken) -> Option<Self> {
52 0 : let mut url = match conf.control_plane_api.as_ref() {
53 0 : Some(u) => u.clone(),
54 0 : None => return None,
55 : };
56 :
57 0 : if let Ok(mut segs) = url.path_segments_mut() {
58 0 : // This ensures that `url` ends with a slash if it doesn't already.
59 0 : // That way, we can subsequently use join() to safely attach extra path elements.
60 0 : segs.pop_if_empty().push("");
61 0 : }
62 :
63 0 : let mut client = reqwest::ClientBuilder::new();
64 :
65 0 : if let Some(jwt) = &conf.control_plane_api_token {
66 0 : let mut headers = reqwest::header::HeaderMap::new();
67 0 : headers.insert(
68 0 : "Authorization",
69 0 : format!("Bearer {}", jwt.get_contents()).parse().unwrap(),
70 0 : );
71 0 : client = client.default_headers(headers);
72 0 : }
73 :
74 0 : Some(Self {
75 0 : http_client: client.build().expect("Failed to construct HTTP client"),
76 0 : base_url: url,
77 0 : node_id: conf.id,
78 0 : cancel: cancel.clone(),
79 0 : })
80 0 : }
81 :
82 0 : async fn retry_http_forever<R, T>(
83 0 : &self,
84 0 : url: &url::Url,
85 0 : request: R,
86 0 : ) -> Result<T, RetryForeverError>
87 0 : where
88 0 : R: Serialize,
89 0 : T: DeserializeOwned,
90 0 : {
91 0 : let res = backoff::retry(
92 0 : || async {
93 0 : let response = self
94 0 : .http_client
95 0 : .post(url.clone())
96 0 : .json(&request)
97 0 : .send()
98 0 : .await?;
99 0 :
100 0 : response.error_for_status_ref()?;
101 0 : response.json::<T>().await
102 0 : },
103 0 : |_| false,
104 0 : 3,
105 0 : u32::MAX,
106 0 : "calling control plane generation validation API",
107 0 : &self.cancel,
108 0 : )
109 0 : .await
110 0 : .ok_or(RetryForeverError::ShuttingDown)?
111 0 : .expect("We retry forever, this should never be reached");
112 0 :
113 0 : Ok(res)
114 0 : }
115 : }
116 :
117 : impl ControlPlaneGenerationsApi for ControlPlaneClient {
118 : /// Block until we get a successful response, or error out if we are shut down
119 0 : async fn re_attach(
120 0 : &self,
121 0 : conf: &PageServerConf,
122 0 : ) -> Result<HashMap<TenantShardId, ReAttachResponseTenant>, RetryForeverError> {
123 0 : let re_attach_path = self
124 0 : .base_url
125 0 : .join("re-attach")
126 0 : .expect("Failed to build re-attach path");
127 0 :
128 0 : // Include registration content in the re-attach request if a metadata file is readable
129 0 : let metadata_path = conf.metadata_path();
130 0 : let register = match tokio::fs::read_to_string(&metadata_path).await {
131 0 : Ok(metadata_str) => match serde_json::from_str::<NodeMetadata>(&metadata_str) {
132 0 : Ok(m) => {
133 0 : // Since we run one time at startup, be generous in our logging and
134 0 : // dump all metadata.
135 0 : tracing::info!(
136 0 : "Loaded node metadata: postgres {}:{}, http {}:{}, other fields: {:?}",
137 : m.postgres_host,
138 : m.postgres_port,
139 : m.http_host,
140 : m.http_port,
141 : m.other
142 : );
143 :
144 0 : Some(NodeRegisterRequest {
145 0 : node_id: conf.id,
146 0 : listen_pg_addr: m.postgres_host,
147 0 : listen_pg_port: m.postgres_port,
148 0 : listen_http_addr: m.http_host,
149 0 : listen_http_port: m.http_port,
150 0 : })
151 : }
152 0 : Err(e) => {
153 0 : tracing::error!("Unreadable metadata in {metadata_path}: {e}");
154 0 : None
155 : }
156 : },
157 0 : Err(e) => {
158 0 : if e.kind() == std::io::ErrorKind::NotFound {
159 : // This is legal: we may have been deployed with some external script
160 : // doing registration for us.
161 0 : tracing::info!("Metadata file not found at {metadata_path}");
162 : } else {
163 0 : on_fatal_io_error(&e, &format!("Loading metadata at {metadata_path}"))
164 : }
165 0 : None
166 : }
167 : };
168 :
169 0 : let request = ReAttachRequest {
170 0 : node_id: self.node_id,
171 0 : register,
172 0 : };
173 :
174 : fail::fail_point!("control-plane-client-re-attach");
175 :
176 0 : let response: ReAttachResponse = self.retry_http_forever(&re_attach_path, request).await?;
177 0 : tracing::info!(
178 0 : "Received re-attach response with {} tenants",
179 0 : response.tenants.len()
180 : );
181 :
182 0 : Ok(response
183 0 : .tenants
184 0 : .into_iter()
185 0 : .map(|rart| (rart.id, rart))
186 0 : .collect::<HashMap<_, _>>())
187 0 : }
188 :
189 : /// Block until we get a successful response, or error out if we are shut down
190 0 : async fn validate(
191 0 : &self,
192 0 : tenants: Vec<(TenantShardId, Generation)>,
193 0 : ) -> Result<HashMap<TenantShardId, bool>, RetryForeverError> {
194 0 : let re_attach_path = self
195 0 : .base_url
196 0 : .join("validate")
197 0 : .expect("Failed to build validate path");
198 0 :
199 0 : let request = ValidateRequest {
200 0 : tenants: tenants
201 0 : .into_iter()
202 0 : .map(|(id, gen)| ValidateRequestTenant {
203 0 : id,
204 0 : gen: gen
205 0 : .into()
206 0 : .expect("Generation should always be valid for a Tenant doing deletions"),
207 0 : })
208 0 : .collect(),
209 0 : };
210 0 :
211 0 : failpoint_support::sleep_millis_async!("control-plane-client-validate-sleep", &self.cancel);
212 0 : if self.cancel.is_cancelled() {
213 0 : return Err(RetryForeverError::ShuttingDown);
214 0 : }
215 :
216 0 : let response: ValidateResponse = self.retry_http_forever(&re_attach_path, request).await?;
217 :
218 0 : Ok(response
219 0 : .tenants
220 0 : .into_iter()
221 0 : .map(|rt| (rt.id, rt.valid))
222 0 : .collect())
223 0 : }
224 : }
|
