Fix flaky issues on MacOS CI (buck2, libzstd) (#200)

Summary:
Here are some flaky issue on MacOS CI that I have seen:

* https://github.com/pytorch/executorch/actions/runs/6053110980/job/16428034274 with `dyld[68891]: Library not loaded: rpath/libzstd.1.dylib` error.  This is the same issue happening on PyTorch, so I copy the fix there in the `print_cmake_info` function
* Buck build fails on MacOS.  I'm not sure the reason here but retry probably works:
  * https://github.com/pytorch/executorch/actions/runs/6029912472/job/16360557094 with `Failed to connect to buck daemon`
  * https://github.com/pytorch/executorch/actions/runs/6052938317/job/16427461951 with some internal errors with `write_file`

Pull Request resolved: #200

Reviewed By: guangy10, dbort

Differential Revision: D48926042

Pulled By: huydhn

fbshipit-source-id: 6e48e670a2989c658f204c76f096cca2c9fe19e6

Author: huydhn

.ci/scripts/setup-macos.sh

@@ -57,10 +57,28 @@ install_buck() {
   fi
 }
 
+# This is the same rpath fix copied from PyTorch macos setup script
+# https://github.com/pytorch/pytorch/blob/main/.ci/pytorch/macos-common.sh
+print_cmake_info() {
+  CMAKE_EXEC=$(which cmake)
+  echo "$CMAKE_EXEC"
+
+  export CMAKE_EXEC
+  # Explicitly add conda env lib folder to cmake rpath to address the flaky issue
+  # where cmake dependencies couldn't be found. This seems to point to how conda
+  # links $CMAKE_EXEC to its package cache when cloning a new environment
+  install_name_tool -add_rpath @executable_path/../lib "${CMAKE_EXEC}" || true
+  # Adding the rpath will invalidate cmake signature, so signing it again here
+  # to trust the executable. EXC_BAD_ACCESS (SIGKILL (Code Signature Invalid))
+  # with an exit code 137 otherwise
+  codesign -f -s - "${CMAKE_EXEC}" || true
+}
+
 # NB: we need buck2 in all cases because cmake build also depends on calling
 # buck2 atm
 install_buck
 install_conda
 install_pip_dependencies
+print_cmake_info
 install_executorch
 build_executorch_runner "${BUILD_TOOL}"

.ci/scripts/utils.sh

@@ -5,6 +5,16 @@
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
+reset_buck() {
+  # On MacOS, buck2 daemon can get into a weird non-responsive state
+  buck2 kill && buck2 clean
+  rm -rf ~/.buck/buckd
+}
+
+retry () {
+    "$@" || (sleep 30 && reset_buck && "$@") || (sleep 60 && reset_buck && "$@")
+}
+
 install_executorch() {
   which pip
   # Install executorch, this assumes that Executorch is checked out in the
@@ -40,8 +50,8 @@ install_pip_dependencies() {
 }
 
 build_executorch_runner_buck2() {
-  # Build executorch runtime
-  buck2 build //examples/executor_runner:executor_runner
+  # Build executorch runtime with retry as this step is flaky on macos CI
+  retry buck2 build //examples/executor_runner:executor_runner
 }
 
 build_executorch_runner_cmake() {
@@ -50,7 +60,9 @@ build_executorch_runner_cmake() {
   rm -rf "${CMAKE_OUTPUT_DIR}" && mkdir "${CMAKE_OUTPUT_DIR}"
 
   pushd "${CMAKE_OUTPUT_DIR}" || return
-  cmake -DBUCK2=buck2 -DPYTHON_EXECUTABLE="${PYTHON_EXECUTABLE}" ..
+  # This command uses buck2 to gather source files and buck2 could crash flakily
+  # on MacOS
+  retry cmake -DBUCK2=buck2 -DPYTHON_EXECUTABLE="${PYTHON_EXECUTABLE}" ..
   popd || return
 
   if [ "$(uname)" == "Darwin" ]; then

examples/custom_ops/test_custom_ops.sh

@@ -11,6 +11,9 @@
 
 set -e
 
+# shellcheck source=/dev/null
+source "$(dirname "${BASH_SOURCE[0]}")/../../.ci/scripts/utils.sh"
+
 test_buck2_custom_op_1() {
   local model_name='custom_ops_1'
   echo "Exporting ${model_name}.pte"
@@ -34,7 +37,7 @@ test_cmake_custom_op_1() {
   (rm -rf cmake-out \
     && mkdir cmake-out \
     && cd cmake-out \
-    && cmake -DBUCK2=buck2 \
+    && retry cmake -DBUCK2=buck2 \
         -DREGISTER_EXAMPLE_CUSTOM_OP=1 \
         -DPYTHON_EXECUTABLE="$PYTHON_EXECUTABLE" ..)
 
@@ -85,7 +88,7 @@ test_cmake_custom_op_2() {
   (rm -rf cmake-out \
     && mkdir cmake-out \
     && cd cmake-out \
-    && cmake -DBUCK2=buck2 \
+    && retry cmake -DBUCK2=buck2 \
       -DREGISTER_EXAMPLE_CUSTOM_OP=2 \
       -DCMAKE_PREFIX_PATH="$CMAKE_PREFIX_PATH" \
       -DPYTHON_EXECUTABLE="$PYTHON_EXECUTABLE" ..)

examples/quantization/test_quantize.sh

@@ -9,6 +9,9 @@
 
 set -e
 
+# shellcheck source=/dev/null
+source "$(dirname "${BASH_SOURCE[0]}")/../../.ci/scripts/utils.sh"
+
 get_shared_lib_ext() {
   UNAME=$(uname)
   if [[ $UNAME == "Darwin" ]];
@@ -47,7 +50,7 @@ test_cmake_quantization() {
   (rm -rf cmake-out \
     && mkdir cmake-out \
     && cd cmake-out \
-    && cmake -DBUCK2=buck2 \
+    && retry cmake -DBUCK2=buck2 \
       -DREGISTER_QUANTIZED_OPS=ON \
       -DCMAKE_PREFIX_PATH="$CMAKE_PREFIX_PATH" \
       -DPYTHON_EXECUTABLE="$PYTHON_EXECUTABLE" ..)

examples/selective_build/test_selective_build.sh

@@ -12,6 +12,9 @@
 # 4. (TODO) Select from a serialized model (.pte)
 set -e
 
+# shellcheck source=/dev/null
+source "$(dirname "${BASH_SOURCE[0]}")/../../.ci/scripts/utils.sh"
+
 test_buck2_select_all_ops() {
     echo "Exporting MobilenetV3"
     ${PYTHON_EXECUTABLE} -m examples.export.export_example --model_name="mv3"
@@ -57,7 +60,7 @@ test_cmake_select_all_ops() {
     (rm -rf cmake-out \
         && mkdir cmake-out \
         && cd cmake-out \
-        && cmake -DBUCK2="$BUCK" \
+        && retry cmake -DBUCK2="$BUCK" \
             -DBUILD_SELECTIVE_BUILD_TEST=ON \
             -DCMAKE_BUILD_TYPE=Release \
             -DSELECT_ALL_OPS=ON \
@@ -81,7 +84,7 @@ test_cmake_select_ops_in_list() {
     (rm -rf cmake-out \
         && mkdir cmake-out \
         && cd cmake-out \
-        && cmake -DBUCK2="$BUCK" \
+        && retry cmake -DBUCK2="$BUCK" \
             -DMAX_KERNEL_NUM=16 \
             -DBUILD_SELECTIVE_BUILD_TEST=ON \
             -DCMAKE_BUILD_TYPE=Release \
@@ -108,7 +111,7 @@ test_cmake_select_ops_in_yaml() {
     (rm -rf cmake-out \
         && mkdir cmake-out \
         && cd cmake-out \
-        && cmake -DBUCK2="$BUCK" \
+        && retry cmake -DBUCK2="$BUCK" \
             -DBUILD_SELECTIVE_BUILD_TEST=ON \
             -DCMAKE_BUILD_TYPE=Release \
             -DSELECT_OPS_YAML=ON \