[Bug 1621629] Re: package-data-downloader fails to process download requests
Brian Murray
brian at ubuntu.com
Mon Sep 12 21:59:49 UTC 2016
Thanks for discovering and working on this issue. This all sounds
reasonable to me. Looking at the update-notifier package I see there is
an automated test for package_data_downloader. Would you mind looking
at writing a test for parts 1 and 3 of the problem you've identified?
Thanks again!
** Changed in: update-notifier (Ubuntu)
Importance: Undecided => High
** Changed in: update-notifier (Ubuntu)
Assignee: (unassigned) => Brian Murray (brian-murray)
--
You received this bug notification because you are a member of Ubuntu
Sponsors Team, which is subscribed to the bug report.
https://bugs.launchpad.net/bugs/1621629
Title:
package-data-downloader fails to process download requests
Status in update-notifier package in Ubuntu:
New
Bug description:
A persistent bug exists in the update notifier that affects the ttf-mscorefonts-installer and may also affect other packages. On boot, a user is presented with an update notification indicating that the ttf-mscorefonts-installer failed to install and needs installation:
---snip---
The following packages requested additional data downloads after package
installation, but the data could not be downloaded or could not be
processed.
ttf-mscorefonts-installer
The download will be attempted again later, or you can try the download
again now. Running this command requires an active Internet connection.
---endsnip---
Running the command fails. Manual installation of the ttf-mscorefonts-installer does not correct the notification. The issue was traced to the /usr/lib/update-notifier/package-data-downloader script.
There are several issues in the package-data-downloader script:
1. The function process_download_requests() checks the hook date against the stamp date for a installation package, such as
Hook File: /usr/share/package-data-downloads/ttf-mscorefonts-installer
Stamp File: /var/lib/update-notifier/package-data-downloads/ttf-mscorefonts-installer
with the code:
---snip---
try:
hook_date = os.stat(file).st_mtime
stamp_date = os.stat(stampfile).st_mtime
if hook_date < stamp_date:
continue
---endsnip---
The stamp file exist if there was a previous installation that did not produce a permanent failure. However, a normal failure sets up the condition to run this script. The failure condition is not cleared on success. The code related to determining success and failure is:
---snip---
result = subprocess.call(command)
if result:
# There's no sense redownloading if the script fails
permanent_failures.append(relfile)
else:
create_or_update_stampfile(stampfile)
---endsnip---
Therefore, in the event of a non-permanent failure, the stamp file is created or updated which prevents an followup download to correct the non-permanent failure.
2. The function process_download_requests() also tests for keywords 'script', 'should-download', 'url', and 'sha256' in the para list from the Hook file to determine action:
---snip---
for para in hook.iter_paragraphs(open(file)):
* . . . . . if 'script' in para:
if not files:
record_failure(relfile)
break
* . . . . . . . command = [para['script']]
* . . . . . . . if 'should-download' in para:
db = debconf.DebconfCommunicator('update-notifier')
try:
* . . . . . . . . . . . should = db.get(para['should-download'])
if should == "false":
# Do nothing with this file.
break
---endsnip---
---snip---
# Not in a 'script' stanza, so we should have some urls
try:
* . . . . . . . files.append(para['url'])
* . . . . . . . sums.append(para['sha256'])
except Exception as e:
---endsnip---
where * note the lines affected. These keywords begin with an uppercase in the related files:
---snip---
Url: http://downloads.sourceforge.net/corefonts/webdin32.exe
Sha256: 64595b5abc1080fba8610c5c34fab5863408e806aafe84653ca8575bed17d75a
Script: /usr/lib/msttcorefonts/update-ms-fonts
Should-Download: msttcorefonts/accepted-mscorefonts-eula
---endsnip---
3. The function process_download_requests() tracks so called "failures" and "permanent_failures". When a failure occurs, the following code sets up the condition for reprocessing later using the function trigger_update_notifier():
---snip---
previous_failures = existing_permanent_failures()
# We only report about "permanent" failures when there are new ones,
# but we want the whole list of permanently-failing hooks so when
# we clobber the update-notifier file we don't lose information the
# user may not have seen yet
if permanent_failures:
new_failures = False
for failure in permanent_failures:
if failure not in previous_failures:
mark_hook_failed(failure, permanent=True)
previous_failures.append(failure)
new_failures = True
if new_failures:
trigger_update_notifier(previous_failures, permanent=True)
# Filter out new failure reports for permanently-failed packages
our_failures = [x for x in failures if x not in previous_failures]
if our_failures:
for failure in our_failures:
mark_hook_failed(failure)
trigger_update_notifier(our_failures)
---endsnip---
SOLUTION:
====================
The existing code set up several conditions based on the existence of the following files:
stamp file
notifier file
notifier permanent file
The conditions are:
EXISTS ( stamp file )
EXISTS ( stamp file AND notifier file )
EXISTS ( notifier file AND notifier permanent file )
EXISTS ( notifier permanent file )
When no permanent failures are found (no notifier permanent file), the
stamp file is set even when normal failures exist (notifier file).
This causes the next script execution to skip to the next install
item. In the event of success, failure conditions are cleared, but
the notifier files are never cleared causing persistent requests to
update that never get fulfilled.
For part 1, adding code handle the failure condition when a stamp file exists would be prudent:
---snip---
try:
> . . . . . if not os.path.exists(NOTIFIER_FILE) and not os.path.exists(NOTIFIER_PERMANENT_FILE):
> . . . . . . . hook_date = os.stat(file).st_mtime
> . . . . . . . stamp_date = os.stat(stampfile).st_mtime
> . . . . . . . if hook_date < stamp_date:
> . . . . . . . . . continue
> . . . . . elif os.path.exists(stampfile):
> . . . . . . . os.unlink(stampfile)
---endsnip---
where > indicated the added/modded code. This corrects the problem by allowing any perceived failures, permanent or not, to reinstall the downloads. It always removes the stampfile so that if a permanent failure develops, it is handled correctly later.
For part 2, the paragraph keyword lookups may or may not care about case, but update the keywords:
'script', 'should-download', 'url', 'sha256'
to:
'Script', 'Should-Download', 'Url', 'Sha256'
so that they match the file.
For part 3, the comment suggests that the so called "update-notifier file" will be clobbered, but it never is. On a non-permament failure, the failed and permanent-failure marker files are removed when function create_or_update_stampfile() is called. The code should be changed to handle cleared failure conditions (i.e., empty failure lists) as follows:
---snip---
previous_failures = existing_permanent_failures()
# We only report about "permanent" failures when there are new ones,
# but we want the whole list of permanently-failing hooks so when
# we clobber the update-notifier file we don't lose information the
# user may not have seen yet
if permanent_failures:
new_failures = False
for failure in permanent_failures:
if failure not in previous_failures:
mark_hook_failed(failure, permanent=True)
previous_failures.append(failure)
new_failures = True
if new_failures:
trigger_update_notifier(previous_failures, permanent=True)
> . if not previous_failures and os.path.exists(NOTIFIER_PERMANENT_FILE):
> . . . os.unlink(NOTIFIER_PERMANENT_FILE)
# Filter out new failure reports for permanently-failed packages
our_failures = [x for x in failures if x not in previous_failures]
if our_failures:
for failure in our_failures:
mark_hook_failed(failure)
trigger_update_notifier(our_failures)
> . elif os.path.exists(NOTIFIER_FILE):
> . . . os.unlink(NOTIFIER_FILE)
---endsnip---
where > indicated the added code.
In total, when no failure is found, the script will create the stamp
file and clear the notifier files. For permanent failures, the stamp
file is not created and any notifier files will persist until the next
attempted update. For normal failures, the stamp and non-permanent
notifier file persists until the next attempted update. On any
update, if notifier files exist, the script will attempt to correct
the install. If just the stamp file exists, then that install is
skipped.
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/update-notifier/+bug/1621629/+subscriptions
More information about the Ubuntu-sponsors
mailing list