Re: svn commit: r775634 - /couchdb/branches/0.9.x/src/couchdb/couch_db.erl

[Antony Blakey]

On 18/05/2009, at 8:44 AM, Adam Kocoloski wrote:
Sure, I agree. Consider this commit a stopgap until we're able to  
pin down the source of the problem.  Antony, any chance you can make  
that 100% reproducible case publicly available?

I'll ask. Apart from being 3.5G it also contains mainly non-public  
work-in-progress client data.

I may be able to make it available selectively with a commercial-in- 
confidence rider. Would that help? My replication is still unreliable  
so I need to track that down, and hence don't have time to fix the  
cause of this bug given that I have this workaround.

BTW: I've just moved to R13B, and I think the problem isn't occurring  
- I'm chasing it down. If that's the case then we can chalk it up to  
an erlang bug.


Adam

On May 17, 2009, at 5:10 PM, Damien Katz wrote:

+%% on rare occasions ibrowse seems to process a chunked response  
incorrectly
+%% and include an extra "\r" in the last chunk.  This code ensures  
that we
+%% truncate the downloaed attachment at the length specified in  
the metadata.

That really sucks. We need to either fix ibrowse or drop it  
altogether.

-Damien

On May 17, 2009, at 8:29 AM, kocol...@apache.org wrote:

Author: kocolosk
Date: Sun May 17 12:29:22 2009
New Revision: 775634

URL: http://svn.apache.org/viewvc?rev=775634&view=rev
Log:
work around ibrowse giving response chunks too many bytes.  Thanks  
Antony.

Modified:
 couchdb/branches/0.9.x/src/couchdb/couch_db.erl

Modified: couchdb/branches/0.9.x/src/couchdb/couch_db.erl
URL: 
http://svn.apache.org/viewvc/couchdb/branches/0.9.x/src/couchdb/couch_db.erl?rev=775634&r1=775633&r2=775634&view=diff
= 
= 
= 
= 
= 
= 
= 
= 
= 
= 
====================================================================
--- couchdb/branches/0.9.x/src/couchdb/couch_db.erl (original)
+++ couchdb/branches/0.9.x/src/couchdb/couch_db.erl Sun May 17  
12:29:22 2009
@@ -649,12 +649,24 @@
  {ok, SpAcc};
write_streamed_attachment(Stream, F, LenLeft, nil) ->
  Bin = F(),
-    {ok, StreamPointer} = couch_stream:write(Stream, Bin),
-    write_streamed_attachment(Stream, F, LenLeft - size(Bin),  
StreamPointer);
+    TruncatedBin = check_bin_length(LenLeft, Bin),
+    {ok, SpAcc} = couch_stream:write(Stream, TruncatedBin),
+    write_streamed_attachment(Stream, F, LenLeft -  
size(TruncatedBin), SpAcc);
write_streamed_attachment(Stream, F, LenLeft, SpAcc) ->
  Bin = F(),
-    {ok, _} = couch_stream:write(Stream, Bin),
-    write_streamed_attachment(Stream, F, LenLeft - size(Bin),  
SpAcc).
+    TruncatedBin = check_bin_length(LenLeft, Bin),
+    {ok, _} = couch_stream:write(Stream, TruncatedBin),
+    write_streamed_attachment(Stream, F, LenLeft -  
size(TruncatedBin), SpAcc).
+
+%% on rare occasions ibrowse seems to process a chunked response  
incorrectly
+%% and include an extra "\r" in the last chunk.  This code  
ensures that we
+%% truncate the downloaed attachment at the length specified in  
the metadata.
+check_bin_length(LenLeft, Bin) when size(Bin) > LenLeft ->
+    <<ValidData:LenLeft/binary, Crap/binary>> = Bin,
+    ?LOG_ERROR("write_streamed_attachment has written too much  
expected: ~p" ++
+        " got: ~p tail: ~p", [LenLeft, size(Bin), Crap]),
+    ValidData;
+check_bin_length(_, Bin) -> Bin.

enum_docs_since_reduce_to_count(Reds) ->
  couch_btree:final_reduce(





Antony Blakey
--------------------------
CTO, Linkuistics Pty Ltd
Ph: 0438 840 787

The project was so plagued by politics and ego that when the engineers  
requested technical oversight, our manager hired a psychologist instead.
  -- Ron Avitzur