66019 - Downloading URL above comes from cache but is incomplete

Status: RESOLVED FIXED

(Core :: Networking: Cache, defect)

Description

[Randell Jesup [:jesup] (needinfo me)]

FreeBSD 4.1 20010118xx

I was downloading the above URL (26MB) and hit an assertion (before selecting a
destination).  after some debugging, I continued, and selected a destination.  I
never saw a "downloading" requester (completion indication).  I went away for a
while, and when I came back a 17MB file was there.  Obviously it was truncated. 
So I deleted it and clicked on the link again.  No downloading requester was
shown, and the 17MB file reappeared immediately (and never grew).

I assume the file is cached, and that the original download timed out or errored
out in some way, and the download isn't being either continued when it's
selected, or it's not being flushed because it was incomplete.

Comment 1

[gordon]

Interesting.  What was your disk cache size set to?

Comment 2

[Randell Jesup [:jesup] (needinfo me)]

5MB (default) (3MB ram cache)

Clearing both memory and disk caches made the problem go away, indicating that
it is a cache problem.

Comment 3

[gordon]

Well, maybe.  The question is what does the 17 meg file have to do with the 
cache.  If the file was trying to get loaded into a 5 meg disk cache, it should 
have failed long before reaching 17 meg.

Are you saying you can successfully download the 26 meg file now?  Or can you 
reproduce the partial 17 meg download?

Comment 4

[Randell Jesup [:jesup] (needinfo me)]

After clearing the caches, I had no problem downloading the 27MB file.

I'm sure the cache is involved, since before clearing it, requesting a download
would cause a 17MB file to appear basically instantly.

I assume mozilla was caching the download as it occured the first time, and when
it failed (for some reason) the cache entry wasn't marked as partial (or dumped,
whatever's appropriate).

Comment 5

[gordon]

Darin, this looks like an HTTP download.  Can you provide some information on how 
HTTP uses the cache for downloads, and how it recovers (or not) from errors?  
It's possible the cache entry got corrupted somehow.

Comment 6

[Darin Fisher]

HTTP does not distinguish downloads from regular GETs w.r.t. how it uses the
cache.  This may change in the future.  As far as error recovery is concerned,
HTTP does very little.  This will change.

Comment 7

[neeti]

I think the TRUNCATED_CONTENT flag is not being set, when the download is 
incomplete. If this flag is set, it causes the data in the cache as invalid and 
issues a new request.

Comment 8

[Darin Fisher]

neeti: sounds correct to me... can you generate a patch?

Comment 9

[neeti]

To generate a patch, I would have to reproduce the bug(I am having a hard time 
doing this). I am able to succesfully download the above 28.7 MB file 
succesfully. I think, when we are having the incomplete downloads, we are either 
not hitting the OnStopRequest() in InterceptStreamListener, or we are hitting 
it, aStatus has a successful value of 0 ( This is the place where the 
TRUNCATED_CONTENT flag is set)

Randell,
Are you able to load this page partially all the time?

Thanks,
Neeti

Comment 10

[Randell Jesup [:jesup] (needinfo me)]

No.  It gave me 17MB once, from then on it gave me 17MB until I cleared the
caches, and since then I haven't had a problem.  I suggest disconnecting your
ethernet during the download until it times out.

I don't know for certain what caused it to be 17MB the first time.
I _assume_ it was a network problem, but it could have been a server bug or a
mozilla bug.

Comment 11

[neeti]

Cache bugs to Gordon

Comment 12

[gordon]

I believe this was a bug in the old cache code which we are no longer using.  
Marking fixed.  If we see it again, we'll open a new bug.