From 4fb2677c4a6849b6f140214fcb84dce2cc6c7899 Mon Sep 17 00:00:00 2001
From: Kyle Brennan <kyle@gitpod.io>
Date: Mon, 25 Sep 2023 13:36:05 -0400
Subject: [PATCH] [content-service] download s3 content using s5cmd (#18783)

* [content-service] download s3 content using s5cmd

Fixes ENG-884

* No pie for you

* [content-service] tune s5cmd based on testing

Results in https://gist.github.com/kylos101/8c49b65d257cf9f642a45877081efc26
---
 components/content-service/pkg/storage/s3.go | 41 ++++++++++++--------
 components/ws-daemon/leeway.Dockerfile       |  6 +++
 2 files changed, 30 insertions(+), 17 deletions(-)

diff --git a/components/content-service/pkg/storage/s3.go b/components/content-service/pkg/storage/s3.go
index 0f36c5b795f773..8641dd9f06a21c 100644
--- a/components/content-service/pkg/storage/s3.go
+++ b/components/content-service/pkg/storage/s3.go
@@ -9,6 +9,7 @@ import (
 	"errors"
 	"fmt"
 	"os"
+	"os/exec"
 	"path/filepath"
 	"strings"
 
@@ -289,33 +290,39 @@ func (s3st *s3Storage) DownloadSnapshot(ctx context.Context, destination string,
 	return s3st.download(ctx, destination, name, mappings)
 }
 
+// download object using s5cmd (prior to which we used aws sdk)
 func (s3st *s3Storage) download(ctx context.Context, destination string, obj string, mappings []archive.IDMapping) (found bool, err error) {
-	downloader := s3manager.NewDownloader(s3st.client, func(d *s3manager.Downloader) {
-		d.Concurrency = defaultCopyConcurrency
-		d.PartSize = defaultPartSize * megabytes
-		d.BufferProvider = s3manager.NewPooledBufferedWriterReadFromProvider(25 * megabytes)
-	})
-
-	s3File, err := os.CreateTemp("", "temporal-s3-file")
+	tempFile, err := os.CreateTemp("", "temporal-s3-file")
 	if err != nil {
 		return true, xerrors.Errorf("creating temporal file: %s", err.Error())
 	}
-	defer os.Remove(s3File.Name())
-
-	_, err = downloader.Download(ctx, s3File, &s3.GetObjectInput{
-		Bucket: aws.String(s3st.Config.Bucket),
-		Key:    aws.String(obj),
-	})
+	tempFile.Close()
+
+	args := []string{
+		"cp",
+		// # of file parts to download at once
+		"--concurrency", "20",
+		// size in MB of each part
+		"--part-size", "25",
+		destination,
+		tempFile.Name(),
+	}
+	cmd := exec.Command("s5cmd", args...)
+	out, err := cmd.CombinedOutput()
 	if err != nil {
-		return false, err
+		log.WithError(err).WithField("out", string(out)).Error("unexpected error downloading file")
+		return true, xerrors.Errorf("unexpected error downloading file")
 	}
 
-	_, err = s3File.Seek(0, 0)
+	tempFile, err = os.Open(tempFile.Name())
 	if err != nil {
-		return false, err
+		return true, xerrors.Errorf("unexpected error opening downloaded file")
 	}
 
-	err = archive.ExtractTarbal(ctx, s3File, destination, archive.WithUIDMapping(mappings), archive.WithGIDMapping(mappings))
+	defer os.Remove(tempFile.Name())
+	defer tempFile.Close()
+
+	err = archive.ExtractTarbal(ctx, tempFile, destination, archive.WithUIDMapping(mappings), archive.WithGIDMapping(mappings))
 	if err != nil {
 		return true, xerrors.Errorf("tar %s: %s", destination, err.Error())
 	}
diff --git a/components/ws-daemon/leeway.Dockerfile b/components/ws-daemon/leeway.Dockerfile
index 243f57a33c23d4..6d08a2b104237b 100644
--- a/components/ws-daemon/leeway.Dockerfile
+++ b/components/ws-daemon/leeway.Dockerfile
@@ -9,6 +9,11 @@ RUN apk add --no-cache curl file \
   && chmod +x runc.amd64 \
   && if ! file runc.amd64 | grep -iq "ELF 64-bit LSB pie executable"; then echo "runc.amd64 is not a binary file"; exit 1;fi
 
+RUN curl -OsSL https://github.com/peak/s5cmd/releases/download/v2.2.2/s5cmd_2.2.2_Linux-64bit.tar.gz \
+ && tar -xzvf s5cmd_2.2.2_Linux-64bit.tar.gz s5cmd \
+ && chmod +x s5cmd \
+ && if ! file s5cmd | grep -iq "ELF 64-bit LSB executable"; then echo "s5cmd is not a binary file"; exit 1;fi
+
 FROM ubuntu:22.04
 
 # trigger manual rebuild increasing the value
@@ -46,6 +51,7 @@ RUN apt update \
     /var/tmp/*
 
 COPY --from=dl /dl/runc.amd64 /usr/bin/runc
+COPY --from=dl /dl/s5cmd /usr/bin/s5cmd
 
 # Add gitpod user for operations (e.g. checkout because of the post-checkout hook!)
 RUN groupadd -r -g 33333 gitpod \