From fda1b150a8eeefbd35d6032bf98f7a2af79fd097 Mon Sep 17 00:00:00 2001
From: Jocelyn Boullier <jocelyn@boullier.bzh>
Date: Sat, 2 Dec 2023 14:42:56 +0100
Subject: [PATCH] feat(gateway): add back retry logic on project error (#1455)

* feat(gateway): attempt at adding back some kind of retry

* refactor: clippy suggestion

---------

Co-authored-by: chesedo <pieter@chesedo.me>
---
 gateway/src/project.rs | 29 +++++++++++++++++++++++++++++
 gateway/src/task.rs    |  6 +++---
 2 files changed, 32 insertions(+), 3 deletions(-)

diff --git a/gateway/src/project.rs b/gateway/src/project.rs
index 538969112..bf0283389 100644
--- a/gateway/src/project.rs
+++ b/gateway/src/project.rs
@@ -663,6 +663,35 @@ where
     }
 }
 
+pub async fn refresh_with_retry(
+    project: Project,
+    ctx: &impl DockerContext,
+) -> Result<Project, Error> {
+    let max_attempt = 3;
+    let mut num_attempt = 1;
+    let mut proj = Box::new(project);
+
+    loop {
+        let refreshed = proj.refresh(ctx).await;
+        match refreshed.as_ref() {
+            Ok(Project::Errored(err)) => match &err.ctx {
+                Some(err_ctx) => {
+                    if num_attempt >= max_attempt {
+                        return refreshed;
+                    } else {
+                        num_attempt += 1;
+                        proj = err_ctx.clone();
+                        tokio::time::sleep(Duration::from_millis(100_u64 * 2_u64.pow(num_attempt)))
+                            .await
+                    }
+                }
+                _ => return refreshed,
+            },
+            _ => return refreshed,
+        }
+    }
+}
+
 #[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
 pub struct ProjectCreating {
     project_name: ProjectName,
diff --git a/gateway/src/task.rs b/gateway/src/task.rs
index e01f5525d..1fc27ca54 100644
--- a/gateway/src/task.rs
+++ b/gateway/src/task.rs
@@ -19,7 +19,7 @@ use uuid::Uuid;
 use crate::project::*;
 use crate::service::{GatewayContext, GatewayService};
 use crate::worker::TaskRouter;
-use crate::{AccountName, Error, ErrorKind, Refresh, State};
+use crate::{AccountName, Error, ErrorKind, State};
 
 // Default maximum _total_ time a task is allowed to run
 pub const DEFAULT_TIMEOUT: Duration = Duration::from_secs(300);
@@ -317,7 +317,7 @@ impl Task<ProjectContext> for RunUntilDone {
 
         // Make sure the project state has not changed from Docker
         // Else we will make assumptions when trying to run next which can cause a failure
-        let project = match ctx.state.refresh(&ctx.gateway).await {
+        let project = match refresh_with_retry(ctx.state, &ctx.gateway).await {
             Ok(project) => project,
             Err(error) => return TaskResult::Err(error),
         };
@@ -351,7 +351,7 @@ impl Task<ProjectContext> for DeleteProject {
     async fn poll(&mut self, ctx: ProjectContext) -> TaskResult<Self::Output, Self::Error> {
         // Make sure the project state has not changed from Docker
         // Else we will make assumptions when trying to run next which can cause a failure
-        let project = match ctx.state.refresh(&ctx.gateway).await {
+        let project = match refresh_with_retry(ctx.state, &ctx.gateway).await {
             Ok(project) => project,
             Err(error) => return TaskResult::Err(error),
         };