- openldap-bugs - openldap.org

Re: (ITS#7775) LMDB terminates Postfix daemon process without logfile record
by wietse＠porcupine.org 05 Jan '14

05 Jan '14

Hallvard Breien Furuseth: > Drafted a branch "mdb/assert" in > <http://folk.uio.no/hbf/OpenLDAP/mdb.git/> and on Ada, based > roughly on Howard's original message to you. I didn't know about > that at first, which has caused some confusion. > > So the callback usually receives an env after all, and a NULL > context since I don't know what Howard intended there. > > Howard, do I push this? Feel free to grab it and tweak a bit:) > > env can be NULL. mdb_env_set_assert(env,cb) also sets a static > variable with the callback for the asserts which do not know of an > env. Could throw away that part later. Currently it only applies > to one assert(). > > The branch also invalidates a transaction on any page allocation > error, since the callers do not always clean up afterwards. Too > aggressive, can be reverted after updating the callers. But for > now it means MDB_MAP_FULL & co won't let the user commit an > inconsistent transaction. Thanks, this is more improvement than I asked for. I don't want to slip my schedule, and to avoid last-minute surprises, I will postpone the review and impact test of these improvements after the stable Postfix release 2-3 weeks from today. Wietse

1 0

Re: (ITS#7775) LMDB terminates Postfix daemon process without logfile record
by h.b.furuseth＠usit.uio.no 05 Jan '14

05 Jan '14

Drafted a branch "mdb/assert" in <http://folk.uio.no/hbf/OpenLDAP/mdb.git/> and on Ada, based roughly on Howard's original message to you. I didn't know about that at first, which has caused some confusion. So the callback usually receives an env after all, and a NULL context since I don't know what Howard intended there. Howard, do I push this? Feel free to grab it and tweak a bit:) env can be NULL. mdb_env_set_assert(env,cb) also sets a static variable with the callback for the asserts which do not know of an env. Could throw away that part later. Currently it only applies to one assert(). The branch also invalidates a transaction on any page allocation error, since the callers do not always clean up afterwards. Too aggressive, can be reverted after updating the callers. But for now it means MDB_MAP_FULL & co won't let the user commit an inconsistent transaction. -- Hallvard

1 0

Re: (ITS#7774) LMDB assertion failure during Postfix cache cleanup
by wietse＠porcupine.org 03 Jan '14

03 Jan '14

Hallvard Breien Furuseth: > I wrote: > >Wietse Venema writes: > >> I can confirm that with LMDB-0.9.10 + ITS#7756 patch, the assertion > >> failure goes away when I close the cursor (and its read transaction) > >> before changing the map or the database, and restore the cursor > >> afterwards. I guess this got broken when things were changed to use > >> external locks in order to avoid world-writable lockfiles. > > Does this mean we can close this ITS? You can close this ITS. Wietse

1 0

Re: (ITS#7775) LMDB terminates Postfix daemon process without logfile record
by wietse＠porcupine.org 03 Jan '14

03 Jan '14

Hallvard Breien Furuseth: > wietse(a)porcupine.org writes: > > Below is my "least painful" patch for both parties. What do you think? > > That's roughly what I wrote was thinking, yes. > > We haven't heard if it will go in at all, though:-) If this ITS > is a reaction to ITS#7774, then the motivation is not catching > an lmdb bug, but letting the user band-aid a user error. Still > makes sense to me, lmdb lets the user screw up in many ways. I agree that we're talking about a stop-gap API in the absence of better error reporting. And I am not talking about users violating the C language model with invalid pointers and such; that would rightfully deserve a core dump. > --- a/libraries/liblmdb/mdb.c > +++ b/libraries/liblmdb/mdb.c Based on visual inspection, that code would work for me. Now the question for me is whether this will be available by the time Postfix 2.11 is released near the end of this month. Code freeze begins in days. I can't distribute LMDB support when daemons fall out of the sky without logfile message, whether it's your mistake, mine, or some middleperson distro maintainer. Prior art: Postfix does not support GnuTLS for the same reason. That would be a support problem. Wietse

1 0

Re: (ITS#7774) LMDB assertion failure during Postfix cache cleanup
by h.b.furuseth＠usit.uio.no 03 Jan '14

03 Jan '14

I wrote: >Wietse Venema writes: >> I can confirm that with LMDB-0.9.10 + ITS#7756 patch, the assertion >> failure goes away when I close the cursor (and its read transaction) >> before changing the map or the database, and restore the cursor >> afterwards. I guess this got broken when things were changed to use >> external locks in order to avoid world-writable lockfiles. Does this mean we can close this ITS? I made a major typo here: > Yes. > > Ending the read-only transaction and starting a new one means you are > no longer seeing the same snapshot, though. So I hope the reason you > couldn't use single transaction which I suggested previously, was **not** > that > the reader transaction needed to keep seeing the same snapshot. -- Hallvard

1 0

Re: (ITS#7775) LMDB terminates Postfix daemon process without logfile record
by h.b.furuseth＠usit.uio.no 03 Jan '14

03 Jan '14

wietse(a)porcupine.org writes: > Below is my "least painful" patch for both parties. What do you think? That's roughly what I wrote was thinking, yes. We haven't heard if it will go in at all, though:-) If this ITS is a reaction to ITS#7774, then the motivation is not catching an lmdb bug, but letting the user band-aid a user error. Still makes sense to me, lmdb lets the user screw up in many ways. Anyway: Why drop the info parameter? If you don't need it after all, that's fine. But then it won't be coming back later. It doesn't fit in the varargs, they're reserved for the message. I ended up declaring a type for it in ldmb.h, to make future expansion a bit friendlier: /** Dummy definition for future extension of #mdb_assert_cb(). */ typedef struct MDB_assertion MDB_assertion; I dislike when a feature makes me #include <stdarg.h>, so I'd not make one myself. I'd build the message in a buffer. Also the code needs fleshing out a bit: Obey NDEBUG, and omit all this if something already #defined assert. See below. Finally, __func__ needs a wrapper. It can go in a renamed "compat" section: --- a/libraries/liblmdb/mdb.c +++ b/libraries/liblmdb/mdb.c @@ -151,3 +150,3 @@ */ -/** @defgroup compat Windows Compatibility Macros +/** @defgroup compat Compatibility Macros * A bunch of macros to minimize the amount of platform-specific ifdefs @@ -158,2 +157,13 @@ */ + + /** Wrapper around __func__, which is a C99 feature */ +#if __STDC_VERSION__ >= 199901L +# define mdb_func_ __func__ +#elif __GNUC__ >= 2 || _MSC_VER >= 1300 +# define mdb_func_ __FUNCTION__ +#else +/* If a debug message says <mdb_unknown>(), update the #if statements above */ +# define mdb_func_ "<mdb_unknown>" +#endif + #ifdef _WIN32 @@ -329,3 +339,3 @@ static txnid_t mdb_debug_start; # define DPRINTF0(fmt, ...) \ - fprintf(stderr, "%s:%d " fmt "\n", __func__, __LINE__, __VA_ARGS__) + fprintf(stderr, "%s:%d " fmt "\n", mdb_func_, __LINE__, __VA_ARGS__) #else Google found more compiler #ifdefs, but I'm not sure which of these are still relevant as non-C99 compilers. Could add them later at need: __func__: (defined(__SUNPRO_C) && defined(__C99FEATURES__)) __FUNC__: __BORLANDC__ >= 0x550 __FUNCTION__: __MWERKS__ >= 0x3000 || defined(__ghs__) || __DMC__ >= 0x810 || __IBMCPP__ >= 500 And the assert() itself: /** assert(3) replacement. * Omitted in programs that #include mdb.c and already define assert(). */ #ifndef assert # ifdef NDEBUG # define assert(expr) ((void) 0) # else # define assert(expr) ((expr) ? (void)0 : \ mdb_assert_fail(#expr, mdb_func_, __FILE__, __LINE__)) static void mdb_assert_fail(const char *expr, const char *func, const char *file, int line) { char buf[400]; sprintf(buf, "%.100s:%d: Assertion '%.200s' failed in %.40s()", file, line, expr, func); if (mdb_assert_cb) mdb_assert_cb(buf, NULL); fprintf(stderr, "%s\n", buf); abort(); } # endif /* NDEBUG */ #endif /* assert */ -- Hallvard

1 0

Re: (ITS#7775) LMDB terminates Postfix daemon process without logfile record
by wietse＠porcupine.org 02 Jan '14

02 Jan '14

Wietse Venema: > Hallvard Breien Furuseth: > > wietse(a)porcupine.org writes: > > >Hallvard Breien Furuseth: > > >> I suppose we could s/assert(foo)/mdb_assert(mc, foo)/ when that > > >> compiles so it can report env, txn, dbi. Howard, should I do that? > > > > > > That was my idea. Allow me to code up an example, it is simpler > > > than arguing in the abstract. > > > > By all means, but I expect we understand each other. I just needed to > > know how muchimpact this needs to have on the lmdb's code. The answer > > is it must walk all over it. And now that I think of it, it'll likely > > introduce merge conflicts in every outstanding branch of any size. > > So my preference is not to do this just yet. > > I am settling for less, and come with a few-line change, so that > LMDB might still end up in this month's stable Postfix release. > > Hang on for a few more minutes. Below is my "least painful" patch for both parties. What do you think? Wietse *** ./work/mdb-mdb/libraries/liblmdb/mdb.c- Tue Nov 12 11:10:33 2013 --- ./work/mdb-mdb/libraries/liblmdb/mdb.c Thu Jan 2 16:30:47 2014 *************** *** 65,71 **** #include <fcntl.h> #endif - #include <assert.h> #include <errno.h> #include <limits.h> #include <stddef.h> --- 65,70 ---- *************** *** 146,151 **** --- 145,155 ---- # error "Two's complement, reasonably sized integer types, please" #endif + /* assert(3) clone. Move this to the project-correct location. */ + static void mdb_assert(const char *, const char *, int, const char *); + #define assert(e) ((e) ? (void) 0 : mdb_assert(__func__, __FILE__, \ + __LINE__, #e)) + /** @defgroup internal MDB Internals * @{ */ *************** *** 3288,3293 **** --- 3292,3311 ---- return MDB_SUCCESS; } + void + (*mdb_assert_cb)(const char *,...); + + static void + mdb_assert(const char *func, const char *file, int line, const char *text) + { + if (mdb_assert_cb != 0) + mdb_assert_cb("Assertion failed: file %s, line %d, function %s: %s", + file, line, func, text); + fprintf(stderr, "Assertion failed: file %s, line %d, function %s: %s\n", + file, line, func, text); + abort(); + } + static int mdb_env_map(MDB_env *env, void *addr, int newsize) { *** ./work/mdb-mdb/libraries/liblmdb/lmdb.h- Tue Nov 12 11:10:33 2013 --- ./work/mdb-mdb/libraries/liblmdb/lmdb.h Thu Jan 2 16:01:37 2014 *************** *** 1429,1434 **** --- 1429,1441 ---- * @return 0 on success, non-zero on failure. */ int mdb_reader_check(MDB_env *env, int *dead); + + /** @brief Notify application of assertion failure. + * This should become obsolete as lmdb's error handling matures. + * @param[in] fmt The assertion message, not including newline. + */ + extern void (*mdb_assert_cb)(const char *fmt, ...); + /** @} */ #ifdef __cplusplus

1 0

Re: (ITS#7775) LMDB terminates Postfix daemon process without logfile record
by wietse＠porcupine.org 02 Jan '14

02 Jan '14

Hallvard Breien Furuseth: > wietse(a)porcupine.org writes: > >Hallvard Breien Furuseth: > >> I suppose we could s/assert(foo)/mdb_assert(mc, foo)/ when that > >> compiles so it can report env, txn, dbi. Howard, should I do that? > > > > That was my idea. Allow me to code up an example, it is simpler > > than arguing in the abstract. > > By all means, but I expect we understand each other. I just needed to > know how muchimpact this needs to have on the lmdb's code. The answer > is it must walk all over it. And now that I think of it, it'll likely > introduce merge conflicts in every outstanding branch of any size. > So my preference is not to do this just yet. I am settling for less, and come with a few-line change, so that LMDB might still end up in this month's stable Postfix release. Hang on for a few more minutes. Wietse

1 0

Re: (ITS#7775) LMDB terminates Postfix daemon process without logfile record
by h.b.furuseth＠usit.uio.no 02 Jan '14

02 Jan '14

wietse(a)porcupine.org writes: >Hallvard Breien Furuseth: >> I suppose we could s/assert(foo)/mdb_assert(mc, foo)/ when that >> compiles so it can report env, txn, dbi. Howard, should I do that? > > That was my idea. Allow me to code up an example, it is simpler > than arguing in the abstract. By all means, but I expect we understand each other. I just needed to know how muchimpact this needs to have on the lmdb's code. The answer is it must walk all over it. And now that I think of it, it'll likely introduce merge conflicts in every outstanding branch of any size. So my preference is not to do this just yet. -- Hallvard

1 0

Re: (ITS#7775) LMDB terminates Postfix daemon process without logfile record
by wietse＠porcupine.org 02 Jan '14

02 Jan '14

Hallvard Breien Furuseth: > wietse(a)porcupine.org writes: > >Hallvard Breien Furuseth: > >> /** User-settable callback for assert(), called before abort() > >> * instead of printing the message. > >> * > >> * assert(some user errors) might be unaffected. This hack > >> * should become obsolete as lmdb's error handling matures. > >> * @param[in] msg The assertion message, not including newline. > >> * @parem[in] info Currently NULL. Might change in the future. > >> */ > >> extern void (*mdb_assert_cb)(char *msg, void *info); > >> > >> (Or maybe call it -before- printing the message, you can abort or > >> longjmp yourself if you don't want the message.) > > > > Looks good. I'll code up an example implementation to set and use > > the callback. The "info" argument should be sufficient to figure > > out which database is failing. > > To repeat, 'info' would be NULL initially. Does that make it useless to > you? Various pieces of code don't know any context of interest to the > user anyway. That is unhelpful, as my opens lots of databases. Instead I propose that for each open database I give you my function pointer AND my context pointer. Then, you call my function with my context pointer when there is a problem. My program opens multiple databases. It is useless if you pass me your LMDB handle, as I now have to find out a) which of all my open databases is LMDB and b) which of those LMDB databases is messed up. That totally breaks all abstractions. > I suppose we could s/assert(foo)/mdb_assert(mc, foo)/ when that > compiles so it can report env, txn, dbi. Howard, should I do that? That was my idea. Allow me to code up an example, it is simpler than arguing in the abstract. Wietse

1 0

Re: (ITS#7774) LMDB assertion failure during Postfix cache cleanup
by h.b.furuseth＠usit.uio.no 02 Jan '14

02 Jan '14

Wietse Venema writes: > As written in the bug report the assertion also failed with > mdb_map_size set to an insanely large value (half a billion) in > which case the map size never changes. Yes... > But, it is good to know that the cursor is invalidated by both map > size changes and transactions that change the database. No, that's the wrong idea. This is a case of a user error breaking snapshot isolation: Ending the transaction invalidates its cursors, and you have to end the transaction in these two cases (modifying when using MDB_NOLOCK, or resizing). Those are "user errors" because that made it simple to code those features. We could add code to make them non-errors on in common cases and on common hosts, but I'm not sure if that's worth the trouble. > I can confirm that with LMDB-0.9.10 + ITS#7756 patch, the assertion > failure goes away when I close the cursor (and its read transaction) > before changing the map or the database, and restore the cursor > afterwards. I guess this got broken when things were changed to use > external locks in order to avoid world-writable lockfiles. Yes. Ending the read-only transaction and starting a new one means you are no longer seeing the same snapshot, though. So I hope the reason you couldn't use single transaction which I suggested previously, was that the reader transaction needed to keep seeing the same snapshot. -- Hallvard

1 0

Re: (ITS#7775) LMDB terminates Postfix daemon process without logfile record
by h.b.furuseth＠usit.uio.no 02 Jan '14

02 Jan '14

wietse(a)porcupine.org writes: >Hallvard Breien Furuseth: >> /** User-settable callback for assert(), called before abort() >> * instead of printing the message. >> * >> * assert(some user errors) might be unaffected. This hack >> * should become obsolete as lmdb's error handling matures. >> * @param[in] msg The assertion message, not including newline. >> * @parem[in] info Currently NULL. Might change in the future. >> */ >> extern void (*mdb_assert_cb)(char *msg, void *info); >> >> (Or maybe call it -before- printing the message, you can abort or >> longjmp yourself if you don't want the message.) > > Looks good. I'll code up an example implementation to set and use > the callback. The "info" argument should be sufficient to figure > out which database is failing. To repeat, 'info' would be NULL initially. Does that make it useless to you? Various pieces of code don't know any context of interest to the user anyway. I suppose we could s/assert(foo)/mdb_assert(mc, foo)/ when that compiles so it can report env, txn, dbi. Howard, should I do that? -- Hallvard

1 0

Re: (ITS#7775) LMDB terminates Postfix daemon process without logfile record
by wietse＠porcupine.org 02 Jan '14

02 Jan '14

Hallvard Breien Furuseth: > > - Invoke an application call-back function with the error code and > > problem description text. If the error code is MDB_PANIC then > > the application knows that it needs to make final arrangements. > > Sounds good to me for now, since it doesn't mean walking all over > the code. mdb.c can just #define its own assert(). No error > code, the context is just "we're about to abort" and I don't know > of other planned callbacks which the code would help. > > /** User-settable callback for assert(), called before abort() > * instead of printing the message. > * > * assert(some user errors) might be unaffected. This hack > * should become obsolete as lmdb's error handling matures. > * @param[in] msg The assertion message, not including newline. > * @parem[in] info Currently NULL. Might change in the future. > */ > extern void (*mdb_assert_cb)(char *msg, void *info); > > (Or maybe call it -before- printing the message, you can abort or > longjmp yourself if you don't want the message.) Looks good. I'll code up an example implementation to set and use the callback. The "info" argument should be sufficient to figure out which database is failing. Wietse

1 0

Re: (ITS#7774) LMDB assertion failure during Postfix cache cleanup
by wietse＠porcupine.org 02 Jan '14

02 Jan '14

Hallvard Breien Furuseth: > Wietse Venema writes: > > Why do you talk about map size changes when I delete a database entry?? > > That's how Postfix uses lmdb. It calls mdb_env_set_mapsize(). Looking As written in the bug report the assertion also failed with mdb_map_size set to an insanely large value (half a billion) in which case the map size never changes. But, it is good to know that the cursor is invalidated by both map size changes and transactions that change the database. I can confirm that with LMDB-0.9.10 + ITS#7756 patch, the assertion failure goes away when I close the cursor (and its read transaction) before changing the map or the database, and restore the cursor afterwards. I guess this got broken when things were changed to use external locks in order to avoid world-writable lockfiles. Wietse

1 0

Re: (ITS#7776) Bulk user add on ldap server
by quanah＠zimbra.com 02 Jan '14

02 Jan '14

--On Thursday, January 02, 2014 9:26 AM +0000 s.puspharaj(a)gmail.com wrote: > Full_Name: Puspharaj > Version: $OpenLDAP: slapd 2.4.23 (Apr 22 2013 05:03:41) > OS: Red Hat Enterprise Linux Server release 6.5 (Santiago) > URL: ftp://ftp.openldap.org/incoming/ > Submission from: (NULL) (103.4.125.30) The ITS system is for reporting bugs, not asking usage questions. I would advise you to take your question to the openldap-technical(a)openldap.org list. I would note a few things: a) The version of OpenLDAP shipped by RHEL is extremely broken. If you insist on using it, you must contact Red Hat for support. b) You can get up to date current OpenLDAP builds from Symas (http://www.symas.com) or the LTB project (<http://ltb-project.org/wiki/download#openldap>) c) I'm going to guess you are using back-bdb or back-hdb. Since you have groups with large membership, it is entirely possible you are running out of locks, lockers, or lock objects. I would advise setting an appropriate debugging level and looking for errors originating from Berkeley DB. --Quanah -- Quanah Gibson-Mount Architect - Server Zimbra, Inc. -------------------- Zimbra :: the leader in open source messaging and collaboration

1 0

(ITS#7776) Bulk user add on ldap server
by s.puspharaj＠gmail.com 02 Jan '14

02 Jan '14

Full_Name: Puspharaj Version: $OpenLDAP: slapd 2.4.23 (Apr 22 2013 05:03:41) OS: Red Hat Enterprise Linux Server release 6.5 (Santiago) URL: ftp://ftp.openldap.org/incoming/ Submission from: (NULL) (103.4.125.30) Hi There, I have around 3000 user in people section were they hold multiple attributes and other things. The problem is, I have couple of groups that holds no user in it, When i tried to add users using ldapmodify command it sometimes not responding or it says the user exist even if the users won't exist in that group. Groupadd.ldif file contains : ################################################################# # dn: cn=typo3_frontend_global_salespartner,ou=Groups,o=company # # changetype: modify # # add: member # # member: uid=alesgian,ou=internal,c=it,ou=People,o=company # # member: uid=alesspin,ou=internal,c=it,ou=People,o=company # ################################################################# Somtimes the uid contains some special characters inbetween like . # , . When the user has the above characters it won't respond or it says the user exist. But when i added manually or with single line of user instead of 1000 user in that file. It accepts the ldapmodify command with that file. The command is ldapmodify -D "cn=Directory Manager,o=company" -f Groupadd.ldif -W This openldap version has any bug related to ldapmodify when adding 2000 users to a group through a single file? Thanks, Puspharaj

1 0

Re: (ITS#7774) LMDB assertion failure during Postfix cache cleanup
by hyc＠symas.com 01 Jan '14

01 Jan '14

Wietse Venema wrote: > Hallvard Breien Furuseth: >>> Sofar the abstraction layer already hides the LMDB-specific MAP_FULL >>> and MAP_RESIZED error conditions. If this abstraction layer needs >>> additional code in order to maintain MDB cursor sanity, then please >>> educate me. >> >> ldmb.h says -- >> >> mdb_env_set_mapsize(): >> "It may be called at later times if no transactions are active in >> this process. Note that the library does not check for this condition, >> the caller must ensure it explicitly." >> MDB_NOLOCK: >> "[caller] must ensure that no readers are using old transactions >> while a writer is active. The simplest approach is to use an >> exclusive lock so that no readers may be active at all when a writer >> begins." >> >> That's why I talked about saving the cursor position and restoring it - >> cursors are per-transaction and you need a new transaction. > > I can remember the last key from mdb_cursor_get() and set the cursor > to that key. There does not appear to be a cursor "save" operation > in the API documentation http://symas.com/mdb/doc/group__mdb.html. Hallvard is suggesting exactly that - remember the key. There is no cursor "save" operation, any saved state would be meaningless after one write transaction. -- -- Howard Chu CTO, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc/ Chief Architect, OpenLDAP http://www.openldap.org/project/

1 0

Re: (ITS#7774) LMDB assertion failure during Postfix cache cleanup
by wietse＠porcupine.org 01 Jan '14

01 Jan '14

Hallvard Breien Furuseth: > > Sofar the abstraction layer already hides the LMDB-specific MAP_FULL > > and MAP_RESIZED error conditions. If this abstraction layer needs > > additional code in order to maintain MDB cursor sanity, then please > > educate me. > > ldmb.h says -- > > mdb_env_set_mapsize(): > "It may be called at later times if no transactions are active in > this process. Note that the library does not check for this condition, > the caller must ensure it explicitly." > MDB_NOLOCK: > "[caller] must ensure that no readers are using old transactions > while a writer is active. The simplest approach is to use an > exclusive lock so that no readers may be active at all when a writer > begins." > > That's why I talked about saving the cursor position and restoring it - > cursors are per-transaction and you need a new transaction. I can remember the last key from mdb_cursor_get() and set the cursor to that key. There does not appear to be a cursor "save" operation in the API documentation http://symas.com/mdb/doc/group__mdb.html. > Maybe you should have a single write-transaction with several cursors, > instead of several transactions. That also cures the "long-lived reader" > problem. That would result in delayed delete operations, and break correctness. If a database entry is updated after it is scheduled for delete, then the entry must not be deleted. Also, I am concerned that would pollute my database-neutral API with LMDB-specific details. Wietse > Howard, maybe NOLOCK should keep the reader table and just drop the > locks? More work though. > > -- > Hallvard

1 0

Re: (ITS#7774) LMDB assertion failure during Postfix cache cleanup
by h.b.furuseth＠usit.uio.no 31 Dec '13

31 Dec '13

Wietse Venema writes: > Why do you talk about map size changes when I delete a database entry?? That's how Postfix uses lmdb. It calls mdb_env_set_mapsize(). Looking closer though, this is in src/util/slmdb.c which is by Howard and you. Sounds like we should wait for whatever Howard has to say. > I need a software abstraction layer between Postfix and LMDB that > provides the following generic interface that is independent of > LMDB internals: > (...snip...) It does, normally. NOLOCK and map resizing interfere, and impose notable limitations. > Sofar the abstraction layer already hides the LMDB-specific MAP_FULL > and MAP_RESIZED error conditions. If this abstraction layer needs > additional code in order to maintain MDB cursor sanity, then please > educate me. ldmb.h says -- mdb_env_set_mapsize(): "It may be called at later times if no transactions are active in this process. Note that the library does not check for this condition, the caller must ensure it explicitly." MDB_NOLOCK: "[caller] must ensure that no readers are using old transactions while a writer is active. The simplest approach is to use an exclusive lock so that no readers may be active at all when a writer begins." That's why I talked about saving the cursor position and restoring it - cursors are per-transaction and you need a new transaction. Maybe you should have a single write-transaction with several cursors, instead of several transactions. That also cures the "long-lived reader" problem. Howard, maybe NOLOCK should keep the reader table and just drop the locks? More work though. -- Hallvard

1 0

Re: (ITS#7774) LMDB assertion failure during Postfix cache cleanup
by wietse＠porcupine.org 31 Dec '13

31 Dec '13

Wietse Venema: > > You're using an old read-only transaction which cannot coexist with: > > - mdb_env_set_mapsize() which moves the map which a cursor in > > the reader is using. > > - several write-transactions + MDB_NOLOCK. The flag means the writers do > > not know about the reader, so they reuse pages from the snapshot the > > reader is using. The reader can survive while the metapages hold > > on to its snapshot, i.e. 1 or 2 write commits (I think). I am willing to be educated. Where in the LMDB API documentation would I have learned that I have to consider these details? Wietse

1 0

Re: (ITS#7774) LMDB assertion failure during Postfix cache cleanup
by wietse＠porcupine.org 31 Dec '13

31 Dec '13

Hallvard Breien Furuseth: > [Due to a typo in your e-mail address, the ITS system did not mail out > this message anywhere. So I'm CC'ing Howard directly, just in case.] Sorry about that. Seems everyone screws up sometimes. > Wietse Venema writes: > > I wrote a test driver that reliably causes LMDB to abort during a > > simulated cache cleanup. This "exploit" produces the same result > > on Linux and FreeBSD, 32-bit and 64-bit systems. > > You're using an old read-only transaction which cannot coexist with: > - mdb_env_set_mapsize() which moves the map which a cursor in > the reader is using. > - several write-transactions + MDB_NOLOCK. The flag means the writers do > not know about the reader, so they reuse pages from the snapshot the > reader is using. The reader can survive while the metapages hold > on to its snapshot, i.e. 1 or 2 write commits (I think). > > I don't know if this is a thinko in your program or miscommunication > between you and Howard about MDB_NOLOCK and mapsize changes. With the > current liblmdb, a map change should involve: Remember the reader's > current position (key), resize the map, renew the txn and cursor, and > reposition the cursor. Why do you talk about map size changes when I delete a database entry?? I need a software abstraction layer between Postfix and LMDB that provides the following generic interface that is independent of LMDB internals: 1) Give me the first or the next element in the database. 2) Delete, look up, update, a specific database entry. 3) Operations under 1) and 2) must be possible to interleave. 4) Locking must be done outside LMDB, because world-writable lock files are not an option. (There is more, but the above is required to use LMDB as a cache with periodic cleanup. I am not makig up stuff here - the above interface has been used in Postfix for many years and it works with all supported databases that have an iterator. Sofar the abstraction layer already hides the LMDB-specific MAP_FULL and MAP_RESIZED error conditions. If this abstraction layer needs additional code in order to maintain MDB cursor sanity, then please educate me. I had expected that LMDB takes care of its cursors itself, since the side effects of other API calls are known inside LMDB only. Wietse > The test works if I (a) turn MDB_NOLOCK into MDB_NOTLS (I know that's > not what you want), and (b) detect map changes in mdb_cursor_get() and > update the cursor to match. > Old 'MDB_val's the reader fetched, are invalid after the mapsize change. > Also, remember that long-lived read-only transactions which write > transactions do know about, prevent them from reusing pages the reader > snapshot is using - resulting in further map growth. > > -- > Hallvard

1 0

Re: (ITS#7775) LMDB terminates Postfix daemon process without logfile record
by h.b.furuseth＠usit.uio.no 31 Dec '13

31 Dec '13

Wietse Venema writes: > (...) > A few alternatives: > > - When a request fails, return a distinct error code such as > MDB_PANIC, and allow the application to look up the text for the > error. For internal errors, yes. It'll take a fair amount of work to get there. And I don't expect ldmb to become enthusiastic at catching user errors, so those may cause crashes or other misbehavior anyway. After all you can disable asserts with NDEBUG, but then you may get e.g. NULL dereferences instead. > - Invoke an application call-back function with the error code and > problem description text. If the error code is MDB_PANIC then > the application knows that it needs to make final arrangements. Sounds good to me for now, since it doesn't mean walking all over the code. mdb.c can just #define its own assert(). No error code, the context is just "we're about to abort" and I don't know of other planned callbacks which the code would help. /** User-settable callback for assert(), called before abort() * instead of printing the message. * * assert(some user errors) might be unaffected. This hack * should become obsolete as lmdb's error handling matures. * @param[in] msg The assertion message, not including newline. * @parem[in] info Currently NULL. Might change in the future. */ extern void (*mdb_assert_cb)(char *msg, void *info); (Or maybe call it -before- printing the message, you can abort or longjmp yourself if you don't want the message.) > The second option preserves 100% backwards compatibility. The first > option may cause functions to return a result that they didn't > return previously. This is about situations where mdb didn't return at all, so compatibility with old behavior will rarely be a problem with new error codes. -- Hallvard

1 0

Re: (ITS#7774) LMDB assertion failure during Postfix cache cleanup
by h.b.furuseth＠usit.uio.no 31 Dec '13

31 Dec '13

[Due to a typo in your e-mail address, the ITS system did not mail out this message anywhere. So I'm CC'ing Howard directly, just in case.] Wietse Venema writes: > I wrote a test driver that reliably causes LMDB to abort during a > simulated cache cleanup. This "exploit" produces the same result > on Linux and FreeBSD, 32-bit and 64-bit systems. You're using an old read-only transaction which cannot coexist with: - mdb_env_set_mapsize() which moves the map which a cursor in the reader is using. - several write-transactions + MDB_NOLOCK. The flag means the writers do not know about the reader, so they reuse pages from the snapshot the reader is using. The reader can survive while the metapages hold on to its snapshot, i.e. 1 or 2 write commits (I think). I don't know if this is a thinko in your program or miscommunication between you and Howard about MDB_NOLOCK and mapsize changes. With the current liblmdb, a map change should involve: Remember the reader's current position (key), resize the map, renew the txn and cursor, and reposition the cursor. The test works if I (a) turn MDB_NOLOCK into MDB_NOTLS (I know that's not what you want), and (b) detect map changes in mdb_cursor_get() and update the cursor to match. Old 'MDB_val's the reader fetched, are invalid after the mapsize change. Also, remember that long-lived read-only transactions which write transactions do know about, prevent them from reusing pages the reader snapshot is using - resulting in further map growth. -- Hallvard

1 0

(ITS#7775) LMDB terminates Postfix daemon process without logfile record
by wietse＠porcupine.org 29 Dec '13

29 Dec '13

Full_Name: Wietse Venema Version: All LMDB versions OS: Linux, BSD URL: ftp://ftp.openldap.org/incoming/ Submission from: (NULL) (70.104.130.26) Subject: LMDB terminates Postfix daemon process without logfile record The LMDB library uses assert() which will abort() the program after writing a message to stderr. In the case of the Postfix MTA, the symptom is a that a daemon process falls out of the sky without logfile record. - This frustrates sysadmins (no logfile message). - This frustrates my efforts to make Postfix automatically recover from non-catastrophic error. In many cases the database is only a cache, a mere optimization. I want the option to just blow it away when LMDB has a problem with it. - Library modules should not abort programs after error. They should return an error result such as EINVAL, or use some other strategy that leaves control with the application. - Writing diagnostics to stderr is not useful for daemon programs in the general case. To fix this problem I suggest that LMDB provide way to notify the application that it was unable to complete a request, so that the application can make proper arrangements. A few alternatives: - When a request fails, return a distinct error code such as MDB_PANIC, and allow the application to look up the text for the error. - Invoke an application call-back function with the error code and problem description text. If the error code is MDB_PANIC then the application knows that it needs to make final arrangements. The second option preserves 100% backwards compatibility. The first option may cause functions to return a result that they didn't return previously. Wietse

1 0

Re: (ITS#7773) slapo-constraint falsely restricts modrdn with count 1
by pierangelo.masarati＠polimi.it 28 Dec '13

28 Dec '13

On 12/28/2013 01:52 PM, michael(a)stroeder.com wrote: > Full_Name: > Version: 2.4.38 > OS: > URL: > Submission from: (NULL) (212.227.35.93) > > > Sometimes it's useful to restrict an existing attribute type declared > multi-valued to a single value with slapo-constraint to avoid modifying the > standard schema. > > Example: > > constraint_attribute > uid > count 1 > restrict="ldap:///ou=example??sub?(objectClass=account)" > > Unfortunately when using uid in the RDN one cannot rename the entry anymore even > when deleteoldrdn is set to true. > > slapo-constraint should handle this case exactly like SINGLE-VALUE RDN > attributes are handled. Should be fixed in git master. Nothing specific to rename; rather, slapo-constraint was not handling internal modifications like softadd/softdel. p. -- Pierangelo Masarati Associate Professor Dipartimento di Scienze e Tecnologie Aerospaziali Politecnico di Milano

1 0

2026

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

openldap-bugs