k8s_cp - wait for remote tar to finish before closing the stream (#1217)

* fix(k8s_cp): wait for remote tar to finish before closing the stream

The module closed the exec connection right after writing the last chunk,
killing tar mid-extract, and never checked its exit status, so a partial
or failed copy was reported as success (#776).

The v4 exec protocol cannot half-close stdin, and busybox tar blocks for
EOF rather than stopping at the end-of-archive marker, so bound the stream
with `head -c N | tar ...`. tar then gets a clean EOF, exits on its own,
and its status is read from the error channel. Containers without /bin/sh
and head keep the previous behaviour and get a warning.

Also add copy_timeout (default 300s, #621) and stream the archive from the
temp file instead of buffering every chunk in memory.

* test(k8s_cp): split combined assert into separate statements

Addresses review feedback on #1217.

* chore(changelog): link the PR and use single-line fragment entries

* fix(linter): reformat with black

* fix(linter): sort imports in test_copy.py

isort --profile black puts ansible_collections in the same third-party
section as pytest, so the blank line between them is a failure.

* fix(k8s_cp): report what tar said when the copy fails early

An early close or a stall is the symptom; tar has usually already
written the cause (a read-only mount, a missing directory) to stderr,
and drain() has it. Discarding it left only the generic truncation or
timeout message.

* refactor(k8s_cp): move the stderr suffix into _fail

The nested fail() helper added a branch inside a closure, which counts
double for nesting and took _stream_tar_to_pod over Sonar's cognitive
complexity limit (14 -> 16). _fail already had the response; give it
the stderr too.

* test(k8s_cp): tempfile for the repeated-copy test directory

Co-authored-by: Bikouo Aubin <79859644+abikouo@users.noreply.github.com>

---------

Co-authored-by: Bikouo Aubin <79859644+abikouo@users.noreply.github.com>
This commit is contained in:
Yuriy Novostavskiy
2026-09-15 09:20:10 +01:00
committed by GitHub
co-authored by Bikouo Aubin
parent 031bc475fa
commit 8a9dd54fe7
8 changed files with 568 additions and 56 deletions
@@ -0,0 +1,8 @@
---
bugfixes:
- k8s_cp - Fix silent truncation when copying to a pod. The module closed the exec connection immediately after writing the last chunk of the tar archive, which killed the remote ``tar`` before it had finished extracting, and never checked the process exit status, so a partial copy was reported as a success. The archive is now bounded with ``head -c`` so ``tar`` gets a clean EOF and exits on its own, and the module waits for that exit and fails on a non-zero status (https://github.com/ansible-collections/kubernetes.core/pull/1217).
minor_changes:
- k8s_cp - Add the ``copy_timeout`` option (default 300 seconds) bounding how long the module will spend streaming an archive to a pod and waiting for the remote ``tar`` to finish, so that a stalled copy fails instead of hanging indefinitely (https://github.com/ansible-collections/kubernetes.core/pull/1217).
- k8s_cp - Stream the tar archive to the pod chunk by chunk instead of building the whole archive in memory first, which cuts peak memory use when copying large files (https://github.com/ansible-collections/kubernetes.core/pull/1217).
- k8s_cp - When copying to a pod, the module now also uses ``/bin/sh`` and ``head`` in the container, when present, to confirm the copy completed. Containers without them keep the previous behaviour and get a warning that completion could not be verified (https://github.com/ansible-collections/kubernetes.core/pull/1217).