From ac4e882a2879d7c33fcc1969523e40a574f4d838 Mon Sep 17 00:00:00 2001
From: "mergify[bot]" <37929162+mergify[bot]@users.noreply.github.com>
Date: Wed, 1 May 2024 15:24:36 +0000
Subject: [PATCH] [8.14](backport #39332) More resilient DRA packaging (#39341)

Occasionally packaging steps from the DRA pipeline may get stuck[^1].
This causes a breach of the global pipeline timeout (currently 1hr) and
cancels the job.

This commit increases the global timeout to 90min, adds one retry per
step and limits the runtime per step to 40min (so that a single stuck
step doesn't exhaust the entire global timeout).

Finally, we shush slack notifications if the retry recovered the step.

In a future PR we will consider also adding a daily DRA build to cover
for cases where the retries didn't help and there were no subsequent
commits to trigger a new build.

[^1]: https://buildkite.com/elastic/beats-packaging-pipeline/builds/114

(cherry picked from commit 726f6e9bdec715f958ba47500e77feb5655b0a48)

---------

Co-authored-by: Dimitrios Liappis <dimitrios.liappis@gmail.com>
---
 .buildkite/packaging.pipeline.yml | 32 +++++++++++++++++++++++++++++++
 1 file changed, 32 insertions(+)

diff --git a/.buildkite/packaging.pipeline.yml b/.buildkite/packaging.pipeline.yml
index c01428100ec..5fd559f458d 100644
--- a/.buildkite/packaging.pipeline.yml
+++ b/.buildkite/packaging.pipeline.yml
@@ -44,6 +44,10 @@ steps:
           provider: gcp
           image: "${IMAGE_UBUNTU_X86_64}"
           machineType: "${GCP_DEFAULT_MACHINE_TYPE}"
+        timeout_in_minutes: 40
+        retry:
+          automatic:
+            - limit: 1
         commands:
           - make build/distributions/dependencies.csv
           - make beats-dashboards
@@ -62,6 +66,10 @@ steps:
           provider: gcp
           image: "${IMAGE_UBUNTU_X86_64}"
           machineType: "${GCP_DEFAULT_MACHINE_TYPE}"
+        timeout_in_minutes: 40
+        retry:
+          automatic:
+            - limit: 1
         commands:
           - make build/distributions/dependencies.csv
           - make beats-dashboards
@@ -86,6 +94,10 @@ steps:
           provider: gcp
           image: "${IMAGE_UBUNTU_X86_64}"
           machineType: "${GCP_DEFAULT_MACHINE_TYPE}"
+        timeout_in_minutes: 40
+        retry:
+          automatic:
+            - limit: 1
         artifact_paths:
           - build/distributions/**/*
         matrix:
@@ -116,6 +128,10 @@ steps:
           provider: "aws"
           imagePrefix: "${AWS_IMAGE_UBUNTU_ARM_64}"
           instanceType: "${AWS_ARM_INSTANCE_TYPE}"
+        timeout_in_minutes: 40
+        retry:
+          automatic:
+            - limit: 1
         artifact_paths:
           - build/distributions/**/*
         matrix:
@@ -142,6 +158,10 @@ steps:
           provider: gcp
           image: "${IMAGE_UBUNTU_X86_64}"
           machineType: "c2-standard-16"
+        timeout_in_minutes: 40
+        retry:
+          automatic:
+            - limit: 1
         artifact_paths:
           - build/distributions/**/*
 
@@ -161,6 +181,10 @@ steps:
           provider: gcp
           image: "${IMAGE_UBUNTU_X86_64}"
           machineType: "${GCP_DEFAULT_MACHINE_TYPE}"
+        timeout_in_minutes: 40
+        retry:
+          automatic:
+            - limit: 1
         artifact_paths:
           - build/distributions/**/*
         matrix:
@@ -191,6 +215,10 @@ steps:
           provider: "aws"
           imagePrefix: "${AWS_IMAGE_UBUNTU_ARM_64}"
           instanceType: "${AWS_ARM_INSTANCE_TYPE}"
+        timeout_in_minutes: 40
+        retry:
+          automatic:
+            - limit: 1
         artifact_paths:
           - build/distributions/**/*
         matrix:
@@ -217,6 +245,10 @@ steps:
           provider: gcp
           image: "${IMAGE_UBUNTU_X86_64}"
           machineType: "c2-standard-16"
+        timeout_in_minutes: 40
+        retry:
+          automatic:
+            - limit: 1
         artifact_paths:
           - build/distributions/**/*