bsdiff logging: distinguish between command timeout and error

[phmccarty]

In the bsdiff logging output, it would be helpful for log messages to distinguish between (a) bsdiff commands that exit with an error (see exit code table below), and (b) bsdiff commands that time out after the fixed 8-minute window. At the moment, both of these conditions log the same error message (Failed to create delta ...).

Here is an easy reference to bsdiff exit codes:

Code	Meaning
0	Success. A delta is created.
1	Created a FULLDL file instead of a delta. The file was (historically) used as a hint to swupd that a "fullfile download" was needed instead of a delta download. In modern times, we delete the file, since swupd knows that the absence of a delta provides the same hint :-)
> 1	Fatal error. At the moment, the only possible value is 255 (returning -1 in the program). Other error codes > 1 but < 255 are reserved for other types of fatal errors in the future.

[reaganlo]

@phmccarty By looking at the code to create deltas, I think there are few more issues here. I've tried to explain them below and provide a suggestion. Please let me know your thoughts: @otaviobp FYI

Current Behavior:	Errors	Write to bsdiffLog	Remove delta file	Return error
1	bsdiff FULLDL	No	Yes	Yes
2	bsdiff Timeout	Yes	Yes	Yes
3	delta > full	Yes	Yes	Yes
4	test bspatch	Yes	No	Yes
5	calculate hash	No	Yes	Yes
6	hash mismatch	No	Yes	Yes

Notice that if there is an error during bspatch (Step 4), the function is returned with an error but the delta file is not deleted.

Also, since we are in the process of implementing logging, going forward there will not be a separate bsdiff log file. So the column "Write to bsdiff log" will become "Write to log" and all the values should be "Yes".

So unless I'm not aware of a specific reason why its being done like this today, I recommend the following:

Suggested Behavior:

	Errors	Write to log	Remove delta file	Return error
1	bsdiff Error	Yes	Yes	Yes
2	bsdiff Timeout	Yes	Yes	Yes
3	delta > full	Yes	Yes	Yes
4	test bspatch	Yes	Yes	Yes
5	calculate hash	Yes	Yes	Yes
6	hash mismatch	Yes	Yes	Yes

[phmccarty]

@reaganlo Thanks for the detailed breakdown. I think the "No" entries in the "Current Behavior" table were oversights, so I also think the three actions should occur for each of these error conditions, as well as for the separate error condition that I identified in the issue description. I will clarify the issue description to make sure that bsdiff exit codes are documented.

[reaganlo]

@ashleshaAtrey @phmccarty Thanks for updating the error codes. So I have updated the Suggested Behavior to catch all bsdiff errors instead of just bsdiff FULLDL. We are planning to have annotations in the new log file:<timestamp>[<loglevel>][<app>] E.g.

[INF][MIXER] [ERR][MIXER] [ERR][BSDIFF] [ERR][DNF] I am assuming it is ok to have a common annotation for [BSDIFF] with the \ explaining what the error is instead of having different annotations like [BSDIFF-TIMEOUT], [BSDIFF], [BSDIFF-FULLDL] . Please let us know otherwise.

[phmccarty]

@reaganlo I am fine with using [BSDIFF], as long as the specific error types are easily greppable from the \<desc>.

[reaganlo]

@phmccarty Cool. Thanks.

[reaganlo]

Hi @phmccarty ,

@otaviobp pointed out that bspatch always returns an error. So maybe it was intentional that the deltafile was not removed if there was a bspatch error.

Here is an e.g. to create the error: https://gist.github.com/otaviobp/79a966bcc749316d4698f8ff5dab4baf

So either we

1) Revert this commit: Remove delta file in case bspatch test fails OR 2) Update the code to not have the bspatch test while creating deltas as it always returns an error.

What do you think?

[phmccarty]

@otaviobp pointed out that bspatch always returns an error. So maybe it was intentional that the deltafile was not removed if there was a bspatch error.

Here is an e.g. to create the error: https://gist.github.com/otaviobp/79a966bcc749316d4698f8ff5dab4baf

That test is not quite correct... Since most content in Clear is owned by root:root, you need to extract the content with sudo tar xf ... and then apply the delta with sudo bspatch .... So, if you repeat the steps by using sudo where necessary, bspatch will succeed.

[reaganlo]

@phmccarty Yes you are correct. Thanks for the clarification.

[reaganlo]

Fixed in v6.2.4