Re: [Ganglia-developers] Defect of "apr_socket_send" usage

[Brad N. <bni...@no...>]

Committed to trunk r1980, proposed for backport to 3.1 branch

>>> On 3/1/2009 at 10:12 PM, in message
<223...@bj...>,
Jerry
<ad...@12...> wrote: 
> Hi Brad,
> 
> As you said, when the number of bytes actually sent does not match
the 
> expected bytes to send, apr_socket_send() automatically does some
setting to 
> force select/poll on the next write call to handle the incomplete
write 
> first. However, for a TCP request, gmond only poll once, then writes
all data 
> to the socket and close the socket before next poll. So I think we
need to 
> resend the miss part immediately for tcp receive channel. I tested my
patch 
> and it works well for my issue.
> 
> Thanks,
> 
> Jerry
> 
> 
> 
> 
> 2009-02-25?23:36:34，"Brad?Nicholes"?<BNI...@no...> ：
>>???apr_socket_send()?returns?the?number?of?bytes?sent?in?the?3rd
>>parameter.??If?the?number?of?bytes?actually?sent?does?not?match?the
>>expected?bytes?to?send,?apr_socket_send()?automatically?sets?the
>>apr_sockopt?to?APR_INCOMPLETE_WRITE.??What?that?does?is?force
>>select/poll?on?the?next?write?call?to?try?to?handle?the?incomplete?write
>>before?actually?writing?any?more?data?to?the?socket.??If?there?is?an
>>error?of?some?sort,?the?return?value?from?apr_socket_send()?will
>>indicate?a?problem.??However,?I?(or?anybody?else)?should?probably?review
>>the?code?in?gmond.c?to?make?sure?that?an?incomplete?write?doesn't
>>actually?cause?further?problems?and?that?the?return?value?from
>>apr_socket_send()?is?actually?being?handled?correctly.
>>
>>Brad
>>
>>
>>>>>?On?2/24/2009?at?6:57?PM,?in?message
>><326...@bj...>,
>>Jerry
>><ad...@12...>?wrote:
>>>?Hi,
>>>?
>>>?Recently,?in?one?of?my?environments,?I?found?"telnet?xxxx?8649"
>>didn't?
>>>?return?a?well-formed?xml?document.?After?some?checking,?I?found?gmond
>>might?
>>>?lose?to?send?out?some?characters.?For?example,?I?got?some?xml
>>document?like?
>>>?"<METRIC?NAME="part_max_used"?VAL="13.8"?TYPE="float"?UNITS="%"
>>TN="55"?
>>>?TMAX="180"?<EXTRA_DATA>".
>>>?
>>>?After?read?the?code,?I?think?usage?of?"apr_socket_send"?in?"gmond.c"
>>has?
>>>?defect.?We?just?check?return?value?of?"apr_socket_send"?but?do?NOT
>>check?how?
>>>?many?characters?have?been?sent.?I?gave?a?patch?in?attachment.?FYI
>>>?
>>>?Thanks,
>>>?
>>>?Jerry
>>>?
>>>?