diff --git a/crates/federate/src/lib.rs b/crates/federate/src/lib.rs
index d3876226f..7cae2ebe9 100644
--- a/crates/federate/src/lib.rs
+++ b/crates/federate/src/lib.rs
@@ -43,7 +43,7 @@ pub struct SendManager {
 }
 
 impl SendManager {
-  pub fn new(opts: Opts, context: FederationConfig<LemmyContext>) -> Self {
+  fn new(opts: Opts, context: FederationConfig<LemmyContext>) -> Self {
     assert!(opts.process_count > 0);
     assert!(opts.process_index > 0);
     assert!(opts.process_index <= opts.process_count);
@@ -61,11 +61,27 @@ impl SendManager {
     }
   }
 
-  pub fn run(mut self) -> CancellableTask {
-    CancellableTask::spawn(WORKER_EXIT_TIMEOUT, move |cancel| async move {
-      self.do_loop(cancel).await?;
-      self.cancel().await?;
-      Ok(())
+  pub fn run(opts: Opts, context: FederationConfig<LemmyContext>) -> CancellableTask {
+    CancellableTask::spawn(WORKER_EXIT_TIMEOUT, move |cancel| {
+      let opts = opts.clone();
+      let context = context.clone();
+      let mut manager = Self::new(opts, context);
+      async move {
+        let result = manager.do_loop(cancel).await;
+        // the loop function will only return if there is (a) an internal error (e.g. db connection
+        // failure) or (b) it was cancelled from outside.
+        if let Err(e) = result {
+          // don't let this error bubble up, just log it, so the below cancel function will run
+          // regardless
+          tracing::error!("SendManager failed: {e}");
+        }
+        // cancel all the dependent workers as well to ensure they don't get orphaned and keep
+        // running.
+        manager.cancel().await?;
+        LemmyResult::Ok(())
+        // if the task was not intentionally cancelled, then this whole lambda will be run again by
+        // CancellableTask after this
+      }
     })
   }
 
@@ -104,14 +120,24 @@ impl SendManager {
             continue;
           }
           // create new worker
-          let instance = instance.clone();
-          let req_data = self.context.to_request_data();
+          let context = self.context.clone();
           let stats_sender = self.stats_sender.clone();
           self.workers.insert(
             instance.id,
-            CancellableTask::spawn(WORKER_EXIT_TIMEOUT, move |stop| async move {
-              InstanceWorker::init_and_loop(instance, req_data, stop, stats_sender).await?;
-              Ok(())
+            CancellableTask::spawn(WORKER_EXIT_TIMEOUT, move |stop| {
+              // if the instance worker ends unexpectedly due to internal/db errors, this lambda is rerun by cancellabletask.
+              let instance = instance.clone();
+              let req_data = context.to_request_data();
+              let stats_sender = stats_sender.clone();
+              async move {
+                InstanceWorker::init_and_loop(
+                  instance,
+                  req_data,
+                  stop,
+                  stats_sender,
+                )
+                .await
+              }
             }),
           );
         } else if !should_federate {
diff --git a/crates/federate/src/util.rs b/crates/federate/src/util.rs
index 02a90dee9..08186f436 100644
--- a/crates/federate/src/util.rs
+++ b/crates/federate/src/util.rs
@@ -17,7 +17,6 @@ use lemmy_db_schema::{
   traits::ApubActor,
   utils::{get_conn, DbPool},
 };
-use lemmy_utils::error::LemmyResult;
 use moka::future::Cache;
 use once_cell::sync::Lazy;
 use reqwest::Url;
@@ -25,7 +24,6 @@ use serde_json::Value;
 use std::{fmt::Debug, future::Future, pin::Pin, sync::Arc, time::Duration};
 use tokio::{task::JoinHandle, time::sleep};
 use tokio_util::sync::CancellationToken;
-use tracing::error;
 
 /// Decrease the delays of the federation queue.
 /// Should only be used for federation tests since it significantly increases CPU and DB load of the
@@ -61,24 +59,31 @@ impl CancellableTask {
   /// spawn a task but with graceful shutdown
   pub fn spawn<F, R: Debug>(
     timeout: Duration,
-    task: impl FnOnce(CancellationToken) -> F + Send + 'static,
+    task: impl Fn(CancellationToken) -> F + Send + 'static,
   ) -> CancellableTask
   where
-    F: Future<Output = LemmyResult<R>> + Send + 'static,
+    F: Future<Output = R> + Send + 'static,
     R: Send + 'static,
   {
     let stop = CancellationToken::new();
     let stop2 = stop.clone();
-    let task: JoinHandle<LemmyResult<R>> = tokio::spawn(task(stop2));
+    let task: JoinHandle<()> = tokio::spawn(async move {
+      loop {
+        let res = task(stop2.clone()).await;
+        if stop2.is_cancelled() {
+          return;
+        } else {
+          tracing::warn!("task exited, restarting: {res:?}");
+        }
+      }
+    });
     let abort = task.abort_handle();
     CancellableTask {
       f: Box::pin(async move {
         stop.cancel();
         tokio::select! {
             r = task => {
-              if let Err(ref e) = r? {
-                error!("CancellableTask threw error: {e}");
-              }
+              r.context("CancellableTask failed to cancel cleanly, returned error")?;
               Ok(())
             },
             _ = sleep(timeout) => {
diff --git a/src/lib.rs b/src/lib.rs
index c2b5e57c2..26740a444 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -210,14 +210,13 @@ pub async fn start_lemmy_server(args: CmdArgs) -> LemmyResult<()> {
     None
   };
   let federate = (!args.disable_activity_sending).then(|| {
-    let task = SendManager::new(
+    SendManager::run(
       Opts {
         process_index: args.federate_process_index,
         process_count: args.federate_process_count,
       },
       federation_config,
-    );
-    task.run()
+    )
   });
   let mut interrupt = tokio::signal::unix::signal(SignalKind::interrupt())?;
   let mut terminate = tokio::signal::unix::signal(SignalKind::terminate())?;