Recursive, parallel crawl

Cargo.lock

@@ -220,6 +220,17 @@ dependencies = [
  "event-listener",
 ]
 
+[[package]]
+name = "async-recursion"
+version = "1.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2cda8f4bcc10624c4e85bc66b3f452cca98cfa5ca002dc83a16aad2367641bea"
+dependencies = [
+ "proc-macro2 1.0.37",
+ "quote 1.0.18",
+ "syn 1.0.92",
+]
+
 [[package]]
 name = "async-trait"
 version = "0.1.53"
@@ -1295,9 +1306,11 @@ name = "lemmy-stats-crawler"
 version = "0.1.0"
 dependencies = [
  "anyhow",
+ "async-recursion",
  "clap",
  "futures",
  "lemmy_api_common",
+ "log",
  "once_cell",
  "reqwest",
  "semver",

Cargo.toml

@@ -15,3 +15,5 @@ clap = "3.1.15"
 semver = "1.0.9"
 once_cell = "1.10.0"
 lemmy_api_common = "0.16.0"
+async-recursion = "1.0.0"
+log = "0.4.17"

src/crawl.rs

@@ -1,56 +1,46 @@
 use crate::REQUEST_TIMEOUT;
-use anyhow::anyhow;
 use anyhow::Error;
+use async_recursion::async_recursion;
+use futures::future::try_join_all;
 use lemmy_api_common::site::GetSiteResponse;
+use log::info;
 use once_cell::sync::Lazy;
 use reqwest::Client;
 use semver::Version;
 use serde::Serialize;
-use std::collections::VecDeque;
+use std::ops::Deref;
+use std::sync::Arc;
+use tokio::sync::Mutex;
 
 static CLIENT: Lazy<Client> = Lazy::new(Client::default);
 
 pub async fn crawl(
     start_instances: Vec<String>,
-    exclude: Vec<String>,
+    exclude_domains: Vec<String>,
     max_depth: i32,
 ) -> Result<(Vec<InstanceDetails>, i32), Error> {
-    let mut pending_instances: VecDeque<CrawlInstance> = start_instances
+    let params = Arc::new(CrawlParams {
-        .iter()
+        min_lemmy_version: min_lemmy_version().await?,
-        .map(|s| CrawlInstance::new(s.to_string(), 0))
+        exclude_domains,
-        .collect();
+        max_depth,
-    let min_lemmy_version = min_lemmy_version().await?;
+    });
-    let mut crawled_instances = vec![];
+    let crawled_instances = Arc::new(Mutex::new(vec![]));
-    let mut instance_details = vec![];
+    let mut jobs = vec![];
-    let mut failed_instances = 0;
+    for domain in start_instances.into_iter() {
-    while let Some(current_instance) = pending_instances.pop_back() {
+        let job = CrawlJob {
-        crawled_instances.push(current_instance.domain.clone());
+            domain,
-        if current_instance.depth > max_depth || exclude.contains(&current_instance.domain) {
+            current_depth: 0,
-            continue;
+            params: params.clone(),
-        }
+            crawled_instances: crawled_instances.clone(),
-        match fetch_instance_details(&current_instance.domain, &min_lemmy_version).await {
+        };
-            Ok(details) => {
+        jobs.push(job.crawl());
-                if let Some(federated) = &details.site_info.federated_instances.as_ref() {
-                    for i in &federated.linked {
-                        let is_in_crawled = crawled_instances.contains(i);
-                        let is_in_pending = pending_instances.iter().any(|p| &p.domain == i);
-                        if !is_in_crawled && !is_in_pending {
-                            let ci = CrawlInstance::new(i.clone(), current_instance.depth + 1);
-                            pending_instances.push_back(ci);
-                        }
-                    }
-                }
-                instance_details.push(details);
-            }
-            Err(e) => {
-                failed_instances += 1;
-                eprintln!("Failed to crawl {}: {}", current_instance.domain, e)
-            }
-        }
     }
 
+    let mut instance_details: Vec<InstanceDetails> =
+        try_join_all(jobs).await?.into_iter().flatten().collect();
+
     // Sort by active monthly users descending
-    instance_details.sort_by_key(|i| {
+    instance_details.sort_unstable_by_key(|i| {
         i.site_info
             .site_view
             .as_ref()
@@ -59,7 +49,7 @@ pub async fn crawl(
     });
     instance_details.reverse();
 
-    Ok((instance_details, failed_instances))
+    Ok((instance_details, 0))
 }
 
 #[derive(Serialize, Debug)]
@@ -68,43 +58,87 @@ pub struct InstanceDetails {
     pub site_info: GetSiteResponse,
 }
 
-struct CrawlInstance {
+struct CrawlParams {
+    min_lemmy_version: Version,
+    exclude_domains: Vec<String>,
+    max_depth: i32,
+}
+
+struct CrawlJob {
     domain: String,
-    depth: i32,
+    current_depth: i32,
+    params: Arc<CrawlParams>,
+    crawled_instances: Arc<Mutex<Vec<String>>>,
 }
 
-impl CrawlInstance {
+impl CrawlJob {
-    pub fn new(domain: String, depth: i32) -> CrawlInstance {
+    #[async_recursion]
-        CrawlInstance { domain, depth }
+    pub async fn crawl(self) -> Result<Vec<InstanceDetails>, Error> {
+        // need to acquire and release mutix before recursing, otherwise it will deadlock
+        {
+            let mut crawled_instances = self.crawled_instances.deref().lock().await;
+            if crawled_instances.contains(&self.domain) {
+                return Ok(vec![]);
+            } else {
+                crawled_instances.push(self.domain.clone());
+            }
+        }
+
+        if self.current_depth > self.params.max_depth
+            || self.params.exclude_domains.contains(&self.domain)
+        {
+            return Ok(vec![]);
+        }
+        info!("Starting crawl for {}", &self.domain);
+
+        let site_info_url = format!("https://{}/api/v3/site", &self.domain);
+        let site_info = CLIENT
+            .get(&site_info_url)
+            .timeout(REQUEST_TIMEOUT)
+            .send()
+            .await
+            .ok();
+
+        if let Some(site_info2) = site_info {
+            let site_info3 = site_info2.json::<GetSiteResponse>().await.ok();
+            if let Some(site_info4) = site_info3 {
+                let version = Version::parse(&site_info4.version).ok();
+
+                if let Some(version) = version {
+                    if version < self.params.min_lemmy_version {
+                        return Ok(vec![]);
+                    }
+                }
+
+                let mut result = vec![];
+                if let Some(federated) = &site_info4.federated_instances {
+                    for domain in federated.linked.iter() {
+                        let crawl_job = CrawlJob {
+                            domain: domain.clone(),
+                            current_depth: self.current_depth + 1,
+                            params: self.params.clone(),
+                            crawled_instances: self.crawled_instances.clone(),
+                        };
+                        result.push(crawl_job.crawl());
+                    }
+                }
+
+                let mut result2: Vec<InstanceDetails> =
+                    try_join_all(result).await?.into_iter().flatten().collect();
+                info!("Successfully finished crawl for {}", &self.domain);
+                result2.push(InstanceDetails {
+                    domain: self.domain,
+                    site_info: site_info4,
+                });
+
+                return Ok(result2);
+            }
+        }
+
+        Ok(vec![])
     }
 }
 
-async fn fetch_instance_details(
-    domain: &str,
-    min_lemmy_version: &Version,
-) -> Result<InstanceDetails, Error> {
-    let client = Client::default();
-
-    let site_info_url = format!("https://{}/api/v3/site", domain);
-    let site_info = client
-        .get(&site_info_url)
-        .timeout(REQUEST_TIMEOUT)
-        .send()
-        .await?
-        .json::<GetSiteResponse>()
-        .await?;
-
-    let version = Version::parse(&site_info.version)?;
-    if &version < min_lemmy_version {
-        return Err(anyhow!("lemmy version is too old ({})", version));
-    }
-
-    Ok(InstanceDetails {
-        domain: domain.to_owned(),
-        site_info,
-    })
-}
-
 /// calculate minimum allowed lemmy version based on current version. in case of current version
 /// 0.16.3, the minimum from this function is 0.15.3. this is to avoid rejecting all instances on
 /// the previous version when a major lemmy release is published.