inevity / cacheboy Goto Github PK

Something I noticed when playing with the mirror.ac.za install of cacheboy-1.5:

If collapsed_forwarding is turned on, then even on "HIT"s in the access.log 
file, objects weren't 
being HIT for some reason. Ie, the access.log entry said HIT, but the returned 
object was a MISS 
IIRC.

This was solved by just turning the collapsed_forwarding option off.

Now, the objects being fetched here are large, rather than the normal use case 
of 
collapsed_forwarding (ie, very frequently requested cachable small objects) 
which others are using 
it for.

1. ./configure --prefix=/usr/local/squid
--enable-storeio=ufs,aufs,coss,null --enable-snmp --enable-epoll
--enable-follow-x-forwarded-for --enable-x-accelerator-vary --with-maxfd=20480
2. make


I expect to end make succesfully, I get this error instead

make[3]: Entering directory `/usr/local/src/CACHEBOY_1.5/app/tcptest'
depbase=`echo tcptest.o | sed 's|[^/]*$|.deps/&|;s|\.o$||'`; \
if gcc -DHAVE_CONFIG_H -I. -I. -I../../include -I. -I../..    -Wall -g -O2
-D_REENTRANT -MT tcptest.o -MD -MP -MF "$depbase.Tpo" -c -o tcptest.o
tcptest.c; \
then mv -f "$depbase.Tpo" "$depbase.Po"; else rm -f "$depbase.Tpo"; exit 1; fi
tcptest.c: In function `main':
tcptest.c:95: error: structure has no member named `sin_len'
tcptest.c:102: error: structure has no member named `sin6_len'
make[3]: *** [tcptest.o] Error 1
make[3]: Leaving directory `/usr/local/src/CACHEBOY_1.5/app/tcptest'
make[2]: *** [all-recursive] Error 1
make[2]: Leaving directory `/usr/local/src/CACHEBOY_1.5/app/tcptest'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/usr/local/src/CACHEBOY_1.5/app'
make: *** [all-recursive] Error 1



RedHat EL 4.3
Linux sq01.dc.dr 2.6.9-42.0.3.EL #1 Mon Sep 25 17:14:19 EDT 2006 i686 i686
i386 GNU/Linux

Hi Guys 

Check the below criteria while retrieving locally HIT files from Cache 
using Cacheboy-1.5.2 on FreeBSD 7.1 
check that this is only an 11M file stored locally on CacheBoy  yet it is
delayed for no reason ... and the squid process doesn't say anything in the
cache.log .....

Regards

fetching
http://au.download.windowsupdate.com/msdownload/update/driver/drvs/2008/02/20102
369_4c2a23f1983ce0ef9f936659679490326992f8bc.cab
--2008-12-03 09:48:41-- 
http://au.download.windowsupdate.com/msdownload/update/driver/drvs/2008/02/20102
369_4c2a23f1983ce0ef9f936659679490326992f8bc.cab
Connecting to 127.0.0.1:3128... connected.
Proxy request sent, awaiting response...
  HTTP/1.0 200 OK
  Date: Fri, 28 Nov 2008 10:34:21 GMT
  Content-Length: 11969534
  Content-Type: application/octet-stream
  ETag: "808957a6fc6dc81:369"
  Last-Modified: Wed, 13 Feb 2008 04:55:27 GMT
  Accept-Ranges: bytes
  Server: Microsoft-IIS/6.0
  X-Powered-By: ASP.NET
  Age: 422060
  X-Cache: HIT from CBSD-Ubi.comnet.net.lb
  Via: 1.1 CBSD-Ubi.comnet.net.lb:3128 (Cacheboy/CACHEBOY_1.5.2)
  Connection: close
Length: 11969534 (11M) [application/octet-stream]
Saving to: `20102369_4c2a23f1983ce0ef9f936659679490326992f8bc.cab'

 0% [>                                                                    
                                               ] 102,400     --.-K/s  eta
2h 16m

What steps will reproduce the problem?
1. ./configure 
2. make
3. make install

What is the expected output? What do you see instead?

error message on make install

make[1]: Entering directory `/home/mtearle/CACHEBOY_1.0/icons'
x - creating lock directory
x - extracting anthony-binhex.gif (binary)
./icons.shar: line 86: uudecode: command not found
restore of anthony-binhex.gif failed
anthony-binhex.gif: MD5 check failed


What version of the product are you using? On what operating system?

Ubuntu Feisty

Please provide any additional information below.

What steps will reproduce the problem?
1. Piano Player
2. Jerk
3. Clearly a homo

What is the expected output? What do you see instead?

Expected outcome is sex with males.

What version of the product are you using? On what operating system?

Adrian v1.

Please provide any additional information below.

Fag!

When building trunk:

./configure --prefix=/usr --enable-gnuregex --enable-async-io --enable-icmp
--enable-useragent-log --enable-snmp --enable-cache-digests
--enable-follow-x-forwarded-for --enable-storeio="ufs,aufs,diskd"
--enable-removal-policies="heap,lru" --with-maxfd=16384
--enable-cache-digests --enable-epoll --disable-ident-lookups
--enable-truncate --exec-prefix=/usr --bindir=/usr/sbin
--libexecdir=/usr/lib/squid --enable-htcp --enable-carp; make

if gcc -DHAVE_CONFIG_H -DDEFAULT_CONFIG_FILE=\"/usr/etc/squid.conf\" -I.
-I. -I../include -I. -I. -I../include -I../include    -Wall -g -O2
-D_REENTRANT -MT net_db.o -MD -MP -MF ".deps/net_db.Tpo" -c -o net_db.o
net_db.c; \
then mv -f ".deps/net_db.Tpo" ".deps/net_db.Po"; else rm -f
".deps/net_db.Tpo"; exit 1; fi
net_db.c: In function `netdbExchangeHandleReply':
net_db.c:565: error: `hdrs_sz' undeclared (first use in this function)
net_db.c:565: error: (Each undeclared identifier is reported only once
net_db.c:565: error: for each function it appears in.)
make[3]: *** [net_db.o] Error 1
make[3]: Leaving directory `/root/cacheboy-read-only/src'
make[2]: *** [all-recursive] Error 1
make[2]: Leaving directory `/root/cacheboy-read-only/src'
make[1]: *** [all] Error 2
make[1]: Leaving directory `/root/cacheboy-read-only/src'
make: *** [all-recursive] Error 1

I've migrated the ident code out of src/ and into a seperate library; 
IPv6'ifying it in the process.

This needs verification!

Details: zph.bratcheda.org

cc  -O2 -fno-strict-aliasing -pipe  -I/usr/local/include
-I/usr/local/include  -I/usr/include -g -D_REENTRANT  -L/usr/local/lib
-L/usr/local/lib -rpath=/usr/lib:/usr/local/lib -L/usr/lib -rdynamic -o
pinger  pinger.o debug.o globals.o -L../lib -L../libcore -L../libmem
-L../libcb -L../libhelper -L../libiapp -L../libhttp -L../libstat
-L../libsqdns -L../libsqinet -lsqinet -lsqdns -lmiscutil -lcore -lmem -lcb
-lhelper -liapp -lhttp -lstat -lrt -pthread -lmd -lm 
../libcore/libcore.a(tools.o)(.text+0x160): In function `getCurrentTime':
/home/adrian/work/cacheboy/ports/ports/www/cacheboy/work/CACHEBOY_1.4/libcore/to
ols.c:35:
multiple definition of `getCurrentTime'
pinger.o(.text+0x90):/home/adrian/work/cacheboy/ports/ports/www/cacheboy/work/CA
CHEBOY_1.4/src/pinger.c:498:
first defined here
/usr/bin/ld: Warning: size of symbol `getCurrentTime' changed from 35 in
pinger.o to 69 in ../libcore/libcore.a(tools.o)
*** Error code 1

wtf?

Apparently the poll() code is busted and doesn't compile in cacheboy-1.4.
I'm sure that also means I didn't finish committing updates to the select() 
variants too.

What steps will reproduce the problem?
1. Compiled under CentOS 5.2
2. Installed it with 1 peer
3. Did normal browsing of websites

What is the expected output? What do you see instead?
No error is expected.  The squid process is not supposed to shutdown, and
reload itself.

What version of the product are you using? On what operating system?
I'm using CacheBoy 1.5.2

Please provide any additional information below.
I'm also running the new Squid 2.7 STABLE 5 without problems using the same
config, and directories for logs, storage, etc.

Regards
HASSAN

A CACHEBOY_PRE trunk Squid running 65536 FDs panics when running through the 
mgr:filedescriptor array, because the array at a certain point past 32768 has 
junk in it.

Its entirely possible stuff isn't being allocated or handled right for >32768 
FDs.

To duplicate: squidclient mgr:filedescriptors ; on a cacheboy squid w/ 65536 
FDs. FreeBSD 
releng_7.

Request on url that match on storeurl rewrite program, if return 302 (or
possibly 303) could create redirection loop, as the the url in question get
cached by cacheboy.

Regards,

P.Y. Adi Prasaja

example:
http://i157.photobucket.com/albums/t72/gbbp/Images/jewel-heart-notes/images/toda
y-tomorrow-forever.gif

debug 61,5
2009/06/02 22:51:54| storeurlStart:
'http://i157.photobucket.com/albums/t72/gbbp/Images/jewel-heart-notes/images/tod
ay-tomorrow-forever.gif'
2009/06/02 22:51:54| storeurlHandleReply:
{http://cdn.photobucket.com/albums/t72/gbbp/Images/jewel-heart-notes/images/toda
y-tomorrow-forever.gif}

store.log when accessing the url
1243954314.396 RELEASE 00 0004010D 90F427DB485C28A54CF21358F136E5A8   ?   
     ?         ?         ? ?/? ?/? ? ?
1243954314.649 RELEASE 00 0004010C B1D7D1B6040C71D88319A6485B0B56F7  200
1243954143        -1 1244054143 x-squid-internal/vary -1/361 GET
http://i157.photobucket.com/albums/t72/gbbp/Images/jewel-heart-note 
s/images/today-tomorrow-forever.gif
1243954314.653 SWAPOUT 00 00040114 DBED62249D43C58F56BC2B1BA2096943  200
1243954314        -1 1244054314 x-squid-internal/vary -1/361 GET
http://i157.photobucket.com/albums/t72/gbbp/Images/jewel-heart-note 
s/images/today-tomorrow-forever.gif

I disabled (NOTYET'ed) the bits in src/client_side.c which implement address 
masking for logging 
purposes.

There's some easy to use masking code now in libsqinet; just fix up the mask 
config parameter to 
be an sqaddr_t and then fix the code!

I don't know which is correct, removing storeurlInit() or adding
storeurlShudown() :-)

This works for me though:

--- main.c      2008-11-08 14:24:03.000000000 +0700
+++ main.c-     2008-11-08 14:23:25.000000000 +0700
@@ -465,6 +465,7 @@
     dnsShutdown();
 #endif
     redirectShutdown();
+    storeurlShutdown();
     locationRewriteShutdown();
     authenticateShutdown();
     externalAclShutdown();

the local_port for the internal DNS sockets is now '0'; i have a feeling
the sqinet conversion may have blotted some call to getsockname() or something.

Check it out!

Traced to this commit:

r13403: Merge r13382 from playpen/ipv6_request

Building CACHEBOY_0.PRE3 results in:

Making all in libmem
make[1]: Entering directory `/root/CACHEBOY_0.PRE3/libmem'
if 
gcc -DHAVE_CONFIG_H -I. -I. -I../include     -Wall -g -O2 -D_REENTRANT -MT 
MemPool.o -MD -MP -MF ".deps/MemPool.Tpo" -c -o MemPool.o MemPool.c; \
then mv -f ".deps/MemPool.Tpo" ".deps/MemPool.Po"; else 
rm -f ".deps/MemPool.Tpo"; exit 1; fi
In file included from MemPool.c:13:
../libcore/debug.h:32: error: syntax error before "va_list"
make[1]: *** [MemPool.o] Error 1
make[1]: Leaving directory `/root/CACHEBOY_0.PRE3/libmem'
make: *** [all-recursive] Error 1

Configure options used:

./configure --prefix=/usr \
--enable-gnuregex --enable-async-io \
--enable-icmp --enable-useragent-log \
--enable-snmp --enable-cache-digests \
--enable-follow-x-forwarded-for \
--enable-storeio="ufs,aufs,diskd" \
--enable-removal-policies="heap,lru" \
--with-maxfd=16384 --enable-cache-digests \
--enable-epoll --disable-ident-lookups \
--enable-truncate --exec-prefix=/usr \
--bindir=/usr/sbin --libexecdir=/usr/lib/squid \
--enable-htcp --enable-carp

The Solaris Event Ports framework lets me submit socket, disk and and timer 
events.
The Squid/Cacheboy codebase isn't (yet!) ready for a cleanly unified comm/disk 
framework but 
there's nothing stopping a comm-only event ports loop from being written.

Write a basic comm-only event port loop for testing!

What steps will reproduce the problem?
1. kernel 2.6.19.3 patched with tproxy2
2. cacheboy-1.6-r13601 compiled with options --enable-linux-tproxy
3.

What is the expected output? What do you see instead?
expected output is no error while compiled. 

What version of the product are you using? On what operating system?
Fedora Core 6, 64-bit OS on 64-bit machine.
cacheboy-1.6-r13601 
tproxy2
iptables-1.3.7


Please provide any additional information below.

compile options:
./configure --sysconfdir=/etc/squid --enable-epoll --enable-snmp --enable-
removal-policies="heap,lru" --enable-storeio="aufs,coss,diskd,null,ufs" --
enable-linux-netfilter --enable-linux-tproxy --with-pthreads --enable-
cachemgr-hostname=localhost --with-maxfd=16384 --enable-err-
languages=English 

error:
tools.o: In function `restoreCapabilities':
/tmp/cacheboy-1.6-r13601/src/tools.c:1132: undefined reference to 
`comm_ips_restoreCapabilities'
../libiapp/libiapp.a(comm_ips_tproxy2.o): In function `comm_ips_bind':
/tmp/cacheboy-1.6-r13601/libiapp/comm_ips_tproxy2.c:49: undefined 
reference to `sqinet_get_v4_addr'
collect2: ld returned 1 exit status
make[3]: *** [squid] Error 1
make[3]: Leaving directory `/tmp/cacheboy-1.6-r13601/src'
make[2]: *** [all-recursive] Error 1
make[2]: Leaving directory `/tmp/cacheboy-1.6-r13601/src'
make[1]: *** [all] Error 2
make[1]: Leaving directory `/tmp/cacheboy-1.6-r13601/src'
make: *** [all-recursive] Error 1
[root@squid cacheboy-1.6-r13601]#

The code in Cacheboy-1.6 to attempt to resolve the first IP in http_port has 
been disabled - the 
code assumes IPv4 only and the ipcache/fqdncache code hasn't yet been updated 
to support both 
IPv4 and IPv6.

The main difference here is that Cacheboy-1.6 -requires- a configured hostname 
or it will 
complain.

Benchmark separating out the 4k page from the mem_node; this should improve 
memory use 
under FreeBSD. (There are other places to worry about memory usage; this isn't 
one.)

Squid-2 and Squid-3 both make crappy assumptions about the type of "client 
instance" used for 
handling IPv4 addresses.

I've just disabled the code entirely in Cacheboy-1.6; so stuff which relies on 
the client instance id 
(ie, the client-db in SNMP) should just be a NULL operation.

A more sensible way of indexing clients is needed - probably by first figuring 
out how to represent 
IPv6 addresses in SNMP and then using some kind of mapping table to integer 
IDs. I dunno yet.

Add option that can FLUSH ipcache ..

squid -k ipcache

:)

1. After recompiling kernel in freebsd 7.1 release with given info in
tproxy.no-ip.org
when moving to port installation by invoking the following command
 make install clean

getting the error
comm.cc:788: error: 'IP_NONLOCALOK' was not declared in this scope

help needed to resolve !!!
any thing else need do let me know

What steps will reproduce the problem?
1.
2.
3.

What is the expected output? What do you see instead?


What version of the product are you using? On what operating system?


Please provide any additional information below.

Large memory objects chew CPU as the storeClientRef() / storeClientCopy()
API uses an offset which is seeked into the memory object each call.

Modify the stmem code to store an "offset" pointer in each mem_node which
indicates the offset in the object which that mem_node begins at.

Modify the store client to keep a reference to the last ref'ed mem_node and
use that as a starting point for the list walk if the required offset is
greater than the beginning offset of that mem_node.

This is a hack but it should substantially reduce the CPU taken on large
memory objects and should suffice until a replacement storage layer is written.

Some of the debug logging of IP addresses has been disabled because, well, I 
was feeling slack 
and wanted to first get the code up and running.

The "cheating" way would be to make an IPv6 version of inet_ntoa() which uses 
thread local 
storage, and then use that inline.

The "non-cheating" way which will work when bits of this code are threaded 
would be to do this:

if (do_debug(type, level)) {
    LOCAL_ARRAY(char, buf, MAX_IPSTRLEN);
    (void) sqinet_ntoa(&inet_thing, buf, MAX_IPSTRLEN, SQADDR_NONE);
    debug(type, level) ("stuff: %s\n", buf);
}

which is slightly dirty and verbose, but will work.

Summary:

  * unify disk code
  * unify ufs/aufs/diskd somewhat
  * parallelize the rebuild process
  * look at (re)using src/disk.c or similar for coalescing async writes for logging

Detail:

The disk and disk storage code is a mess. ufs, aufs and diskd all share quite a 
large amount of 
storedir code which should be unified somewhat. The original 
sync-but-almost-async stuff is 
still in src/disk.c and used in a few places.

Older Squid releases (up to 2.2?) shoehorned in asyncio support by overloading 
the file_* calls to 
be async! This had the advantage of allowing logging to also be async.

Ideally, the intersection between the asyncio and src/disk.c should be taken to 
implement an 
async disk layer. The "sync" versions of these routines would operate a lot 
like the Squid-3 
AsyncCalls layer is trying to achieve elsewhere, except we can be reasonably 
certain here that 
callers are expecting the callback to be fired at a later date rather than 
called immediately.

The FQDN cache code currently doesn't support IPv6 and so the FQDN IP logging 
code has been 
disabled. Any attempt to use the client FQDN logging field for an access.log 
file will return an IP 
address.

(I'm not sure whether the FQDN code is doing a background lookup or whether it 
blocks waiting for 
the lookup to occur. I should investigate that.)

Hi 

seems you eliminated src/debug.c where the -l localN options are defined

since this seems to be the only way getting squid logging to syslogd 
instead to cache.log I like to ask you considering it again 

thanks, Michel

The referer and useragent logging in src/client_side.c has been disabled for 
now. The problem is 
that both bits of invocation code use fqdnFromAddr() which is only IPv4-aware 
at the present time.

The solution may just be to log the IP and not the FQDN for now?

The squid IPC code uses fd_open() to open FDs returned by the 
pipe/socketpair/fifo syscalls but 
then uses comm_close() on them.

This works fine in Squid-2 (and Squid-3?) but it means I can't plonk down a 
seperate comm and fd 
context without some reworking. It specifically breaks at the moment with 
sqinet - 
sqinet_init(&fd_table[fd].local_address) isn't called but comm_close() calls 
sqinet_done(...) ; and this 
breaks the assumption that you only deinit what you've init'ed!

Anyway. Its been worked around in ipc.c and I've done a cursory inspection to 
make sure other fd's 
aren't fd_open() but comm_close()'ed. It should just be fixed at some point to 
be consistent so 
comm and disk can be properly split from the fd_table[].

Insaling cacheboy 1.6 via ports FreeBSD 7.1-STABLE

Making all in libsqname

 cc -DHAVE_CONFIG_H -I. -I../include    -I/usr/include  -O2
 -fno-strict-aliasing -pipe  -I/usr/include -g -D_REENTRANT -MT 
 ipcache.o -MD -MP -MF .deps/ipcache.Tpo -c -o ipcache.o ipcache.c

 ipcache.c: In function 'ipcacheParse':

 ipcache.c:258: error: 'DNS_INBUF_SZ' undeclared (first use in this 
 function)

 ipcache.c:258: error: (Each undeclared identifier is reported only 
 once
 ipcache.c:258: error: for each function it appears in.)
 *** Error code 1
 Stop in /usr/ports/www/cacheboy16/work/cacheboy-1.6-r13601/libsqname.
 *** Error code 1
 Stop in /usr/ports/www/cacheboy16/work/cacheboy-1.6-r13601.
 *** Error code 1
 Stop in /usr/ports/www/cacheboy16.
 *** Error code 1
 Stop in /usr/ports/www/cacheboy16.

CacheDigest.o(.text+0x163): In function `cacheDigestCreate':
/root/CACHEBOY_PRE/src/CacheDigest.c:72: undefined reference to `memAlloc'
mem.o(.text+0x5ac): In function `memInit':
/root/CACHEBOY_PRE/src/mem.c:261: undefined reference to `peerDigestInitMem'
collect2: ld returned 1 exit status
make[3]: *** [squid] Error 1
make[3]: Leaving directory `/root/CACHEBOY_PRE/src'
make[2]: *** [all-recursive] Error 1
make[2]: Leaving directory `/root/CACHEBOY_PRE/src'
make[1]: *** [all] Error 2
make[1]: Leaving directory `/root/CACHEBOY_PRE/src'
make: *** [all-recursive] Error 1

Cacheboy from SVN crashes almost immediately after startup.

What steps will reproduce the problem?
1. Recompile kernel without IPv6 support
2. Attempt to compile CACHEBOY
3. Will fail

What is the expected output? What do you see instead?
I'd like to compile CACHEBOY without IPv6 support.

What version of the product are you using? On what operating system?
Tested on CentOS 5.2 - 2.6.25.10 Kernel
Tested using CACHEBOY1.5 to 1.6PRE2 (I have resorted back to 1.4.1 for now)

Please provide any additional information below.
Perhaps have it detect the status of IPv6, or support a --disable-ipv6 
option.

Polymix-4 is showing this during a run, even during the early ramp-up phase:

001.22| Xaction.cc:72: error: 2/2 (c14) premature end of msg body
1216833537.616818# obj: 
http://w1141.h1128o1101s1010.bench.tst/w12261d6a.48331351:7ffffdc7/t01/_00000103
.png 
flags: basic,GET, size: 403/3346 xact: 12261d6a.48331351:00001494
[no data to dump]

Squid-2.7 and Squid-2.HEAD aren't showing this problem.

$ squid -k reconfigure
$ dmesg
squid[14648]: segfault at 61206170 eip 4cf3b6bc esp bfa2fde8 error 4

gdb output:
(gdb) bt
#0  0x4cf3b6bc in ?? ()
#1  0x080cf12f in splay_splay (data=0x61206174, top=0x85918a84,
compare=0x85918a84) at splay.c:102
#2  0x080cb34b in authBasicFreeConfig (scheme=0x8a5ca90) at
basic/auth_basic.c:239
#3  0x080cb43f in authenticateBasicFixErrorHeader
(auth_user_request=0x8222ae4, rep=0x8222a80,
     type=HDR_ACCEPT, request=0x82229d8) at basic/auth_basic.c:228
#4  0x080cfe97 in _db_init (options=0x8222ad8 "\005") at
/usr/include/stdlib.h:286
#5  0x080d110a in stringAppend (s=0x82229d8, str=0x8222a44
"�*\"\bX*\"\b\030+\"\b", len=0)
     at String.c:121
#6  0x0804dadc in aclDestroyAcls (head=0x82199b0) at acl.c:2610
#7  0x080555e0 in ?? ()
#8  0x0805d3cc in configDoConfigure () at cache_cf.c:551
#9  0x080961dc in mainInitialize () at main.c:518
#10 0x08096b2a in main (argc=3, argv=0xbfd5b374) at main.c:767

I use CACHEBOY-1.3.2, my squid.conf attached.

thx

Regards,

P.Y. Adi Prasaja

What steps will reproduce the problem?
1. ./configure --enable-ssl --with-openssl=/usr/kerberos
2. make


What is the expected output? What do you see instead?
- configure runs without errors, and my system don´t have any openssl
install.. and configure pass without WARNING, but make report erros with
make ...

What version of the product are you using? On what operating system?
CacheBoy 1.5.2 Stable running on CentOS 5.2 64 bits

After daemon exit (because it´s crashed) .. after updated installation with
patched sources.. i´m getting many of this


2009/01/15 03:27:22| storeClientReadHeader: no URL!
2009/01/15 03:27:22| storeClientReadHeader: no URL!
2009/01/15 03:27:22| storeClientReadHeader: no URL!

There's been a drop in "performance" between 2.7 and 2.HEAD, seen by last.fm.
This is a placeholder for Tony to drop in some further information.

The load balancers may be interpreting the client-side write() split as
"replies taking longer". Implement a patch to 2.7 which splits the header
and  initial body write() into two halves to test whether this change is
responsible for the reduction in throughput.

(And if it is, begin including the writev() and cork() logic..)