From 1f3872007244df93c82238354b913cba7c71bc89 Mon Sep 17 00:00:00 2001
From: Daniel Langbein <daniel@systemli.org>
Date: Thu, 12 Jan 2023 20:37:42 +0100
Subject: [PATCH] improved active receive

---
 src/p1st/btrfs_receive_chunks_active.py | 10 ++++++++--
 src/p1st/repeat.py                      | 12 +++++++++++-
 2 files changed, 19 insertions(+), 3 deletions(-)

diff --git a/src/p1st/btrfs_receive_chunks_active.py b/src/p1st/btrfs_receive_chunks_active.py
index 0420b3d..1d89904 100644
--- a/src/p1st/btrfs_receive_chunks_active.py
+++ b/src/p1st/btrfs_receive_chunks_active.py
@@ -35,6 +35,7 @@ def main():
             inform_cmd = ['ssh',
                           ssh_source,
                           f'cat {shlex.quote(str(inform_path))} && rm {shlex.quote(str(inform_path))}']
+            messages = [Message1.OK.value, Message1.EOF.value]
             rsync_cmd = ['rsync',
                          f'{ssh_source}:{source_chunk_path}',
                          f'{target_chunk_path}']
@@ -43,8 +44,13 @@ def main():
                            f"printf '{Message2.OK_STR.value}' | nc -U {shlex.quote(str(in_socket))}"]
 
             # Wait until next chunk can be transferred from sending side.
-            messages = [Message1.OK.value, Message1.EOF.value]
-            msg = repeat_until_successful(inform_cmd, user_socket)
+            #
+            # It can happen,that the sending side has not yet saved the next chunk.
+            # In that case `cat ...` will fail.
+            # We will retry 3 times, each time after 5 seconds sleep.
+            #
+            # If all 4 `cat ...` attempts failed, we wait until the user has inspected the problem.
+            msg = repeat_until_successful(inform_cmd, user_socket, retries=3, retry_delay_seconds=5)
             if msg not in messages:
                 raise ValueError(f'Invalid message: {msg}')
             last_chunk = msg == Message1.EOF.value
diff --git a/src/p1st/repeat.py b/src/p1st/repeat.py
index b2e2a0c..bd51a7f 100644
--- a/src/p1st/repeat.py
+++ b/src/p1st/repeat.py
@@ -1,3 +1,4 @@
+import time
 from pathlib import Path
 
 from p1st.exec_print_capture import execute_print_capture
@@ -5,7 +6,9 @@ from p1st.unix_sock_input import wait_for_message
 
 
 def repeat_until_successful(command: list[str],
-                            socket_file: Path) -> str:
+                            socket_file: Path,
+                            retries: int = 0,
+                            retry_delay_seconds: float = 5) -> str:
     """
     Executes the given `command`.
 
@@ -17,6 +20,7 @@ def repeat_until_successful(command: list[str],
         out: str
         returncode, out, _err = execute_print_capture(command)
         if returncode == 0:
+            time.sleep(retry_delay_seconds)
             return out
 
         print(f'\n'
@@ -24,6 +28,12 @@ def repeat_until_successful(command: list[str],
               f'\t{command}\n'
               f'\tFor details, see above output.')
 
+        if retries > 0:
+            print('\tRetrying the failed command.'
+                  '\n')
+            retries = retries - 1
+            continue
+
         print(f'Info:\n'
               f'\tPlease fix the above error first. Then continue here:\n'
               f'\tsudo pacman -S --needed openbsd-netcat\n'