build_rootfs.sh: Make debootstrap process more resilient to network glitches

When building Rock Pi images on CloudTop, it was common for debootstrap
to fail to pull down one or more (sometimes many) packages during the
"Retrieving Packages" sequence, amusingly due to network anomalies:

  E: Couldn't download packages: libc-l10n libc6 locales libgmp10 [SNIPPED]
  Filesystem generation process failed.

This resulted in complete failure and required the whole build process
to be started over from scratch, multiple times.

Simply re-invoking the same debootstrap command a few times
successively without cleaning, consistently resulted in all of the
packages being downloaded.

Signed-off-by: Lee Jones <lee.jones@linaro.org>
Change-Id: Iafbf7f55a6f6245911c308c44c8532244522d9d9

1 files changed, 11 insertions, 2 deletions

diff --git a/net/test/build_rootfs.sh b/net/test/build_rootfs.sh
index 6cfac70..8acc47a 100755
--- a/net/test/build_rootfs.sh
+++ b/net/test/build_rootfs.sh
@@ -178,8 +178,17 @@ sudo chown root:root "${workdir}"
 
 # Run the debootstrap first
 cd "${workdir}"
-sudo debootstrap --arch="${arch}" --variant=minbase --include="${packages}" \
-                 --foreign "${suite%-*}" . "${mirror}"
+
+retries=5
+while ! sudo debootstrap --arch="${arch}" --variant=minbase --include="${packages}" \
+        --foreign "${suite%-*}" . "${mirror}"; do
+    retries=$((${retries} - 1))
+    if [ ${retries} -le 0 ]; then
+	failure
+	exit 1
+    fi
+    echo "debootstrap failed - trying again - ${retries} retries left"
+done
 
 # Copy some bootstrapping scripts into the rootfs
 sudo cp -a "${SCRIPT_DIR}"/rootfs/*.sh root/