rizqiv2/vSphere-Backup-Manager
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: improve task waiting robustness, add on-the-fly SHA-256 calculation to downloads, and implement snapshot removal retries

Browse Source
This commit is contained in:
Rizqi 2026-06-22 03:14:06 +07:00
parent e15b64790f
commit 9d5dc0420d
2 changed files with 43 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
backup_core.py
View File

@ -115,12 +115,14 @@ def list_vms(host, user, password, no_verify_ssl=False):
def wait_for_task(task, action_name='job'): def wait_for_task(task, action_name='job'):
while task.info.state == vim.TaskInfo.State.running: while task.info.state not in (vim.TaskInfo.State.success, vim.TaskInfo.State.error):
time.sleep(1) time.sleep(1)
if task.info.state == vim.TaskInfo.State.success: if task.info.state == vim.TaskInfo.State.success:
return task.info.result return task.info.result
else: else:
raise Exception(f"{action_name} did not complete successfully: {task.info.error}") err = task.info.error
err_msg = getattr(err, 'msg', None) or str(err)
raise Exception(f"{action_name} did not complete successfully: {err_msg}")
def create_snapshot(vm, snap_name, desc="backup snapshot", memory=False, quiesce=False): def create_snapshot(vm, snap_name, desc="backup snapshot", memory=False, quiesce=False):
@ -141,7 +143,7 @@ def find_datacenter_for_datastore(content, datastore_name):
def download_datastore_file(host, dc_name, datastore_name, ds_path, local_path, def download_datastore_file(host, dc_name, datastore_name, ds_path, local_path,
session_cookie, verify_ssl=True, progress_cb=None): session_cookie, verify_ssl=True, progress_cb=None):
"""Download a file from a vSphere datastore. progress_cb(bytes_done, bytes_total) is optional.""" """Download a file from a vSphere datastore and return its SHA-256 checksum. progress_cb(bytes_done, bytes_total) is optional."""
# Keep slashes unencoded (safe='/') — vCenter's /folder/ API requires them in the URL path. # Keep slashes unencoded (safe='/') — vCenter's /folder/ API requires them in the URL path.
encoded_path = urllib.parse.quote(ds_path, safe='/') encoded_path = urllib.parse.quote(ds_path, safe='/')
url = (f"https://{host}/folder/{encoded_path}" url = (f"https://{host}/folder/{encoded_path}"
@ -149,6 +151,7 @@ def download_datastore_file(host, dc_name, datastore_name, ds_path, local_path,
headers = {"Cookie": f"vmware_soap_session={session_cookie}"} headers = {"Cookie": f"vmware_soap_session={session_cookie}"}
print(f"Downloading {ds_path} from datastore {datastore_name} to {local_path}") print(f"Downloading {ds_path} from datastore {datastore_name} to {local_path}")
print(f" URL: {url}") print(f" URL: {url}")
sha256 = hashlib.sha256()
with requests.get(url, headers=headers, stream=True, verify=verify_ssl, proxies={"http": None, "https": None}) as r: with requests.get(url, headers=headers, stream=True, verify=verify_ssl, proxies={"http": None, "https": None}) as r:
r.raise_for_status() r.raise_for_status()
total_bytes = int(r.headers.get('Content-Length', 0)) total_bytes = int(r.headers.get('Content-Length', 0))
@ -159,10 +162,12 @@ def download_datastore_file(host, dc_name, datastore_name, ds_path, local_path,
for chunk in r.iter_content(chunk_size=4 * 1024 * 1024): for chunk in r.iter_content(chunk_size=4 * 1024 * 1024):
if chunk: if chunk:
f.write(chunk) f.write(chunk)
sha256.update(chunk)
done_bytes += len(chunk) done_bytes += len(chunk)
if progress_cb: if progress_cb:
progress_cb(done_bytes, total_bytes) progress_cb(done_bytes, total_bytes)
print(f"Download completed ({done_bytes // (1024*1024)} MB)") print(f"Download completed ({done_bytes // (1024*1024)} MB)")
return sha256.hexdigest()
def extract_session_cookie(si): def extract_session_cookie(si):
@ -208,9 +213,20 @@ def find_snapshot_by_name(snapshots, name):
def remove_snapshot(snapshot_obj): def remove_snapshot(snapshot_obj):
print("Removing snapshot") print("Removing snapshot")
task = snapshot_obj.RemoveSnapshot_Task(removeChildren=False) max_retries = 3
wait_for_task(task, 'RemoveSnapshot') for attempt in range(max_retries):
print("Snapshot removed") try:
task = snapshot_obj.RemoveSnapshot_Task(removeChildren=False)
wait_for_task(task, 'RemoveSnapshot')
print("Snapshot removed")
return
except Exception as e:
print(f"Attempt {attempt+1} to remove snapshot failed: {e}")
if attempt < max_retries - 1:
print("Waiting 5 seconds before retrying...")
time.sleep(5)
else:
raise e
def get_file_sha256(filepath, decompress_if_zst=False): def get_file_sha256(filepath, decompress_if_zst=False):
@ -465,7 +481,7 @@ def _run_backup_impl(host, user, password, vm_name, dest, compress, no_verify_ss
_prog('downloading', file_base_pct, _prog('downloading', file_base_pct,
f'Starting file {file_idx+1}/{total_files}: {os.path.basename(ds_path)}') f'Starting file {file_idx+1}/{total_files}: {os.path.basename(ds_path)}')
download_datastore_file( file_sha = download_datastore_file(
host, dc_name, ds_name, ds_path, local_file, session_cookie, host, dc_name, ds_name, ds_path, local_file, session_cookie,
verify_ssl=not no_verify_ssl, verify_ssl=not no_verify_ssl,
progress_cb=make_dl_cb(file_idx, total_files, file_base_pct, progress_cb=make_dl_cb(file_idx, total_files, file_base_pct,
@ -473,13 +489,9 @@ def _run_backup_impl(host, user, password, vm_name, dest, compress, no_verify_ss
) )
downloaded_files.append(local_file) downloaded_files.append(local_file)
# Compute checksum immediately after download # Checksum was computed on-the-fly during download
_prog('downloading', file_base_pct + int(file_share * 0.95), f'Calculating checksum for {os.path.basename(ds_path)}…')
print(f"Calculating SHA-256 checksum for {local_file}")
t0 = time.time()
file_sha = get_file_sha256(local_file)
file_size = os.path.getsize(local_file) file_size = os.path.getsize(local_file)
print(f"SHA-256: {file_sha} (size: {file_size} bytes, took {time.time() - t0:.2f}s)") print(f"SHA-256 (computed on-the-fly): {file_sha} (size: {file_size} bytes)")
# Relative path from dest directory using forward slashes (e.g. "datastore1/Nakivo/Nakivo.vmdk") # Relative path from dest directory using forward slashes (e.g. "datastore1/Nakivo/Nakivo.vmdk")
rel_path = os.path.relpath(local_file, dest).replace(os.sep, '/') rel_path = os.path.relpath(local_file, dest).replace(os.sep, '/')

23
vsphere_backup.py
View File

@ -70,12 +70,14 @@ def find_vm_by_name(content, vm_name):
def wait_for_task(task, action_name='job'): def wait_for_task(task, action_name='job'):
while task.info.state == vim.TaskInfo.State.running: while task.info.state not in (vim.TaskInfo.State.success, vim.TaskInfo.State.error):
time.sleep(1) time.sleep(1)
if task.info.state == vim.TaskInfo.State.success: if task.info.state == vim.TaskInfo.State.success:
return task.info.result return task.info.result
else: else:
raise Exception(f"{action_name} did not complete successfully: {task.info.error}") err = task.info.error
err_msg = getattr(err, 'msg', None) or str(err)
raise Exception(f"{action_name} did not complete successfully: {err_msg}")
def create_snapshot(vm, snap_name, desc="backup snapshot", memory=False, quiesce=False): def create_snapshot(vm, snap_name, desc="backup snapshot", memory=False, quiesce=False):
@ -157,9 +159,20 @@ def find_snapshot_by_name(snapshots, name):
def remove_snapshot(snapshot_obj): def remove_snapshot(snapshot_obj):
print("Removing snapshot") print("Removing snapshot")
task = snapshot_obj.RemoveSnapshot_Task(removeChildren=False) max_retries = 3
wait_for_task(task, 'RemoveSnapshot') for attempt in range(max_retries):
print("Snapshot removed") try:
task = snapshot_obj.RemoveSnapshot_Task(removeChildren=False)
wait_for_task(task, 'RemoveSnapshot')
print("Snapshot removed")
return
except Exception as e:
print(f"Attempt {attempt+1} to remove snapshot failed: {e}")
if attempt < max_retries - 1:
print("Waiting 5 seconds before retrying...")
time.sleep(5)
else:
raise e
def upload_via_sftp(host, user, password, key_filename, local_path, remote_dir): def upload_via_sftp(host, user, password, key_filename, local_path, remote_dir):