Parent Directory
| 
Revision Log
| Links to HEAD: | (view) (annotate) |
| Sticky Revision: |
backport scoreboard-lb_score-2.2.x.patch
* Merge r602542, r603237, r603502, r603543, r604447, r604449, r605314, r605838 from trunk: * Enable the proxy to keep connections persistent in the HTTPS case. Basicly the persistence is created by keeping the conn_rec structure created for our backend connection (whether http or https) in the connection pool. This required to adjust scoreboard.c in a way that its functions can properly deal with a NULL scoreboard handle by ignoring the call or returning an error code. * Use a separate subpool to manage the data for the socket and the connection member of the proxy_conn_rec struct as we destroy this data more frequently than other data in the proxy_conn_rec struct like hostname and addr (at least in the case where we have keepalive connections that timed out and were closed by the backend). This fixes a memory leak with short lived and broken connections. * Fix another memory leak related to PR 44026. Now that we keep the connection data structure alive in the reslist, the live time of c->pool is too long. r->pool has the correct live time since rp dies before r. * Do not register connection_cleanup as cleanup for the conn->pool. In the past it was needed to register connection_cleanup as a cleanup for the frontend connection memory pool (c->pool) to ensure that connection returns into the connection pool if the memory pool of the frontend connection memory pool gets destroyed / cleared. Now we ensure explicitly the connection returns to the connection pool once we finished handling the request. * Tag the pools appropriately to ease memory debugging. * Only sent a flush bucket down the chain if buckets where sent down the chain before that could still be buffered in the network filter. This is the case if we have sent an EOS bucket or if we actually sent buckets with data down the chain. In all other cases we either have not sent any buckets at all down the chain or we only sent meta buckets that are not EOS buckets down the chain. The only meta bucket that remains in this case is the flush bucket which would have removed all possibly buffered buckets in the network filter. If we sent a flush bucket in the case where not ANY buckets were sent down the chain, we break error handling which happens AFTER us. * Using the reslist pool for the proxy_conn_rec structure introduces a memory leak when connections get created and destroyed frequently by the reslist (e.g. destroying idle elements of the reslist). So use the subpool dedicated for the proxy_conn_rec structure to allocate the memory for the structure itself. PR: 44026, 44543 Submitted by: rpluem Reviewed by: jim, rpluem, fielding
Merge 590641 from trunk: Add in SeeRequestTail directive, to handle the shortcoming of only storing 63 bytes of the request, when the requests are longer than that and only vary towards the end; eg: GET /disk1/storage/apache/htdocs/images/image-store1/food/fruits/seeded/apples.jpg GET /disk1/storage/apache/htdocs/images/image-store1/food/fruits/seeded/pears.jpg GET /disk1/storage/apache/htdocs/images/image-store1/food/fruits/seeded/plums.jpg Submitted by: jim Reviewed by: jim, rpluem, niq
Merge r578620 from trunk: PR 40037: try to avoid risk of misleading the lusers when they can't initialise scoreboard after a crash. Submitted by: niq Reviewed by: niq, rpluem, trawick
update license header text
Revert r395231 from the 2.2.x branch. This gets us back to the old place with regard to the copyright statements.
Update the last year of copyright for the 2.2.x branch
No functional change: remove trailing whitespace. This also means that "blank" lines, which had consisted of just spaces and/or tabs are now truly blank lines
Recreate 2.2.x branch from trunk.
Fix some symbols which should -not- be exported, and decorate real_exit_code with ap_ (this symbol must be exported.) If the others were desired to be public, they should have been ap_xxx decorated.
Update copyright year to 2005 and standardize on current copyright owner line.
Keep track of time taken to process requests again.
general property cleanup
veto and revert win64 patch: 64bit changes must percolate from the bottom (APR/system) up -- we can't give the client a 64bit API and then cast it to 32bits internally without introducing security holes on other platforms.
WIN64: API changes to clean up Windows 64bit compile warnings
* server/scoreboard.c (create_namebased_scoreboard): Use apr_shm_remove() rather than the unsufficient apr_file_remove(). PR: 27189
Worker shared data has to be visible across all child processes, so remove creating that data on per-child basis.
Add load balancer support to the scoreboard in preparation for load balancing support in mod_proxy. PR: Obtained from: Submitted by: Mladen Turk <mturk@apache.org> Reviewed by:
fix name of The Apache Software Foundation
fix copyright dates according to the first check in
apply Apache License, Version 2.0
update license to 2004.
On failures, try to add a hint what went wrong
finished that boring job: update license to 2003. Happy New Year! ;-))
stop using APLOG_NOERRNO in calls to ap_log_?error()
Fix a potential size mismatch (apr_off_t might be 64bits on some platforms).
The old, legacy (and unused) code in which the scoreboard was totally and completely contained in a file (SCOREBOARD_FILE) has been removed. This does not affect scoreboards which are *mapped* to files using named-shared-memory at all. This implies that scoreboards must be based, at some level, on native shared memory (mmap, shm_open, shmget, whatever), but the code has assumed that for quite awhile now. Having the scoreboard be *based* on a file makes no sense today. PR: Obtained from: Submitted by: Reviewed by:
ap_create_scoreboard() now initializes the scoreboard if it already exists, so MPMs shouldn't call it for a graceful restart
The source of some ills ... there is no reason to keep going if the scoreboard exists. I suspect this should be a general cleanup as well [at the end of ap_create_scoreboard.] But calling ap_run_pre_mpm with the process->pool should take care of a clobbered scoreboard shm on graceful restart.
Eliminate potential ap_server_root_relative segfaults, with the input of Jeff Trawick's style changes to the first patches. Doesn't include the fixes to ssl [more complex], and we won't trap errors that involve ap_serverroot, since we presume that was normalized on the way in. Therefore, testing ap_server_root_relative(DEFAULT_FOO) cases should never become necessary.
Update our copyright for this year.
Fix the calculation of thread_num in the worker score structure. Submitted by: Stas Bekman Reviewed by: Jeff Trawick
ap_calc_scoreboard_size() needs to be exported for win32
remove a tab
Placing the thread-limit and server-limit values, those which define the size of internal structures in the scoreboard, actually into the scoreboard allow third-party applications to access those structures. Submitted by: Adam Sussman <myddryn@vishnu.vidya.com> Reviewed by: Aaron Bannert
This fixes a bug in mod_status on Windows where restart time was not properly initialized. Move ap_restart_time into the scoreboard global area so the child process on non-forking platforms can have access to it.
Implement new ScoreBoardFile directive logic. This affects how we
create the scoreboard's shared memory segment. We now have the best of
both worlds:
if config specifies ScoreBoardFile
create name-based shared memory, errors are fatal
else /* we get to choose */
create anonymous shared memory
if ENOTIMPL
create name-based shared memory from DEFAULT_SCOREBOARD
else
errors are fatal
This gives us the flexibility to have anonymous shared memory (on platforms
that support it) as well as name-based shared memory when third-party
programs want access to our scoreboard.
The ap_scoreboard_fname static variable is now owned by the scoreboard.c
file, and no longer by the MPMs. The MPMs MUST NOT set ap_scoreboard_fname
to a default, since that will override the default creation logic and
only allow name-based segments.
Submitted by: Aaron Bannert
Reviewed by: Justin Erenkrantz
Moved the #endif of APR_HAS_SHARED_MEMORY to include a reference to the global variable ap_scoreboard_shm. Since this variable was already being excluded by APR_HAS_SHARED_MEMORY at the beginning of the file.
Create the scoreboard (in the parent) in a global pool context, so it survives graceful restarts. This fixes a SEGV during graceful restarts. Children who attach to this scoreboard keep the same pool as before (pchild) since they should detach/unmap when the child process exits.
Change the Windows MPM to only use the pre_mpm phase in the parent process. The child processes use the child_init phase to reattach to the shared memory. This makes Windows work like Unix, which should make it easier for module authors to write portable modules.
fix the problem where a scoreboard init failure could leave mod_cgid stranded a pre_mpm hook can now return failures, so problems in ap_create_scoreboard percolate back to a place where Apache can exit cleanly
axe an unused variable don't pass uninitialized rv to ap_log_error as the error code
No. I don't like this patch. I like what it does [shared scoreboard for Windows, finally] but not how it's implemented. However it works, and is equally crufty to what already exists for win32 listeners and other exposed data within scoreboard.c. To do this right, we need to drop all the external references to data within the scoreboard, and add an accessor for remaining bits (such as Win32's need for direct access to the apr_shm_t.) And within Win32, we need to stack all this _within_ the pre-mpm hook. But those are missions in and of themselves.
A quick hack around Aaron's patch. Win32 'supports' anon shm, but Apache's implementation requires fork().
Just a style update before I work on this thing any more. Removed tabs and lined up parameters and indentation.
Although this patch is technically correct, I'm not happy with the way it gets things done. OTOH, it is a simple enough change to get things working correctly for now. I will come up with the right way to do this in the next couple days. This patch re-enables the use of anonymous shared memory in the scoreboard on platforms that have it.
Added APR_HAS_SHARED_MEMORY to a section of code where it was missing, for those of us that don't have shared memory support.
This patch eliminated from the _SHARED_ segment of the scoreboard all pointer math. This is required for portable scoreboards. vhost becomes the 'vhost name string' so it now survives ap_generation clicks. next was apparently never used. This patch also accounts for the changes to the apr_shm api, and gives Win32 the magic of a shared scoreboard. Breakage aplenty on non-win32 platforms, I suspect, but this radical surgery, and culling of unused functions, was really, really needed.
Make calc_scoreboard_size() and init_scoreboard() usable from MPMs.
fix a horrible bug which caused scoreboard initialation to always exit
The scoreboard is no longer created while we can write to stderr, so call ap_log_error() to note problems allocating the scoreboard. apr_shm_malloc() doesn't set any kind of error code, so don't try to reference errno. setup_shared() now returns an error code to its caller rather than exiting directly. This isn't a complete fix, as ap_create_scoreboard() needs to do the same thing. Currently, when we can't allocate the scoreboard we leave the mod_cgid daemon process stranded.
Clean up a compiler emit for signedness
Change core code to allow an MPM to set hard thread/server limits at startup.
Provide an accessor function for the global_score portion of the scoreboard. Submitted by: Harrie Hazewinkel <harrie@covalent.net> Reviewed by: Aaron Bannert
Export ap_get_parent_scoreboard and ap_get_servers_scoreboard to be usable on Win32 and friends. Submitted by: Harrie Hazewinkel <harrie@covalent.net> Reviewed by: Aaron Bannert
prevent seg faults in mod_status trying to access vhost structures from the former generation, after a restart. It happens mostly when the MPM has threads.
when we bail out due to a fatal error in apr_shm_init(), be sure to print the apr_status_t in numeric form the old message lt-httpd: could not open(create) scoreboard: Unknown resolver error wasn't so useful, but the new, improved message lt-httpd: could not open(create) scoreboard: (22528)Unknown resolver error makes it look like a syscall is returning ENOSPC and that apr shmem is returning (APR_OS_START_SYSERR + errno) instead of simply errno. off to APR land...
adjust to apr_uri_ rename
adjust to UNP_ -> APR_URI_UNP_ rename
clean up the scoreboard properly after threaded processes die gracefully. find_child_by_pid() failed to do its job, because ap_max_daemons_used was too low. There was some recent breakage in threaded that contributes to the problem. However, find_child_by_pid() should not be using the current number of daemons. It could have been recently reduced, which makes it fail to find the exiting process.
Exported symbols for mod_pop
prevent seg faults in mod_status with ExtendedStatus enabled, after a restart. A pointer to the previous generation's vhost server_rec could be left in the worker_score under certain conditions.
Fix ap_create_scoreboard fooness on Win32.
Make scoreboard creation a hook. This allows management modules to have access to the scoreboard at the time that it is created, and at every restart request. Submitted by: Cody Sherr <csherr@covalent.net> Reviewed by: Ryan Bloom
Changed AP_MPMQ_MAX_DAEMONS to refer to MaxClients and added an AP_MPMQ_MAX_DAEMON_USED to refer to the highest daemon index actually used in the scoreboard. I also updated the pertinent calls. Paul J. Reder
Remove a couple fields from the scoreboard that aren't currently used. If we need these, they should be added when we begin to use them.
scratch an old itch - give lingering close its own state in the scoreboard. clean up SERVER_ACCEPTING and SERVER_QUEUEING (never set) while I'm at it.
Add two functions to allow modules to access random parts of the scoreboard. This allows modules compiled for one MPM to access the scoreboard, even if it the server was compiled for another MPM. Submitted by: Harrie Hazewinkel <harrie@covalent.net>
Moved util_uri to apr-util/uri/apr_uri, which means adding the apr_ prefix to all of the uri functions (yuck), changing some includes, and using APR error codes instead of HTTP-specific error codes. Other notes to test this patch: - You need to delete the util_uri.h file - exports picks up on this. - I'd like to remove the apr_uri.h from httpd.h, but that might increase the complexity of this patch even further. Once this patch is accepted (in some form), then I can focus on removing apr_uri.h from httpd.h entirely. I need baby steps (heh) right now. - I imagine that this might break a bunch of stuff in Win32 or other OS builds with foreign dependency files. Any help here is appreciated. This is a start... -- justin Submitted by: Justin Erenkrantz Reviewed by: Roy Fielding
Make first phase changes to the scoreboard data structures in preparation for the rewriting of the scoreboard per my posted design notes. [Paul J. Reder]
fix some generation logic/displays in mod_status by setting the generation field in parent_score
minor scoreboard/status improvements: . get the SS field in extended status output formatted correctly (seconds since beginning of request shouldn't be a huge negative number :) ) . use APR_OS_PID_T_FMT and pid_t where appropriate in mod_status to avoid casting, some of which may have been broken on some architectures
tweak ap_get_remote_host() so that the caller can find out if she got back an IP address mod_access needed to know this, but the old code didn't handle IPv6
Allow modules to query the MPM about it's execution profile. This query API can and should be extended in the future, but for now, max_daemons, and threading or forking is a very good start. Non-Unix MPM's do have the MPM query function, although there is no garauntee that the information is perfect, please check. Submitted by: Jon Travis <jtravis@covalent.net>
*) fix inline handling. we had: apr_inline, APR_INLINE, USE_GNU_INLINE, and INLINE. Now, we just have APR_INLINE and APR_HAS_INLINE. - convert all usage - note that apr_general messed up the defn (compared to apr.h) - simplify the inline decision logic in os/*/os.h - simplify the code in os/*/os-inline.c *) toss ap_checkconv() [no longer used]
Update copyright to 2001
Start getting extended status working again.
*) include the strings stuff the right way (some platforms use string.h; apr_want figures all that out for us) *) include ap_mpm.h to pick up a moved func declaration
Fix a warning.
Clean up some of the includes: - explicitly include apr_lib.h since ap_config.h doesn't - use apr_want.h where possible - use APR_HAVE_ where possible - remove some unneeded includes
*) remove obsolete http_conf_globals.h *) simplify all this #ifdef'ing: just nuke the unixd.h header. it isn't actually needed.
BeOS doesn't have unixd.h either.
Replace reinit_scoreboard with ap_create_scoreboard.
renaming various functions for consistency sake see: http://apr.apache.org/~dougm/apr_rename.pl PR: Obtained from: Submitted by: Reviewed by:
Fix graceful shutdown with the scoreboard changes. Basically, we want to set the sb_type only after we zero out the scoreboard. Before this, we were segfaulting when we tried to free() a shared memory scoreboard.
Get the Unix MPMs restarting again. If we are going to register a cleanup for ap_cleanup_scoreboard, then we have to kill that cleanup with the same function. This also makes ap_cleanup_scoreboard a non-static function, and makes ap_cleanup_shared_mem a static function.
Setup the server start time correctly. This also gets the server uptime to be reported correctly. I have also moved this code into a common location, so that individual MPMs do not need to worry about setting up the start time.
Get OS/2 MPM running again after the reunification of the scoreboard code. mod_status still doesn't work right, I'll look into that next.
cleanups need to return apr_status_t
Ooops, that S and D are right next to each other.
Begin restructuring scoreboard code to enable adding back in the ability to use IPC other than shared memory. Get mod_status working on Windows again. Still to do: Rename some of the function APIs. Replace all calls to reinit_scoreboard with ap_create_scoreboard. Add back in support for scoreboard files.
Bring mod_status for 2.0 back in line with mod_status for 1.3. This is basically a straight port of the 1.3 module to 2.0. The MPMs need to be modified a bit to work with mod_status, but prefork, mpmt_pthread, and dexter have already been changed. I will fix perchild tonight. There is a lot of common code that can be abstracted, and there seems to be a small bug with regard to what mpmt_pthread and dexter report as current connections. ExtendedStatus does work again, although until the bug mentioned above is fixed, it isn't as useful on mpmt_pthread and dexter. Next week, I will look at allowing other modules to add data to the STATUS page and possibly to the scoreboard itself.
This form allows you to request diffs between any two revisions of this file. For each of the two "sides" of the diff, enter a numeric revision.
| apache@apache.org | ViewVC Help |
| Powered by ViewVC 1.1.2 |