[MonetDB-users] Serious concurrency issues via python api
I have multiple processes reading and writing to the same table and I've noticed really persistent concurrency problems:
File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line 208, in execute self.__store_result(block) File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line 520, in __store_result self.__exception_handler(InterfaceError, "Unknown state, %s" % block) File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line 574, in __exception_handler raise exception_class(message) InterfaceError: Unknown state, &2 65541 -1 !ERROR: COMMIT: transaction is aborted because of concurency conflicts, will ROLLBACK instead
I'm not sure what the problem is, but it exists in such a way that all operations that load or update information in MonetDB have to be put into a loop to continuously retry.
I'm using autocommit, but I would expect MonetDB to handle concurrency issues in a decent way, like maybe blocking a query until the other query finished. I'm not clobbering data, most of these are just bulk load operations that insert new rows.
Is there any way to address this issue without having to putting retry loops around all of the code I'm writing to write to the database?
73, Matthew W. Jones (KI4ZIB) http://matburt.net
Dear Matthew,
A quick initial reaction. MonetDB uses an optimistic concurrency control scheme, which is geared at high-read, low-write applications. If you hit the system with multiple concurrent writes on the same table, then indeed to secure acid properties, transactions are bound to fail. MonetDB indeed conforms to the semantics of a transaction safe system.
Blocking queries is equivalent to relation level locking. This is not the preferred way to go for the kernel, because someone will immediately ask for more granularities and als relaxation of the ACID principles.
There are several application scenarios to deal with this issue, but all depend on the semantics required.
regards, Martin
Matthew Jones wrote:
I have multiple processes reading and writing to the same table and I've noticed really persistent concurrency problems:
File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line 208, in execute self.__store_result(block) File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line 520, in __store_result self.__exception_handler(InterfaceError, "Unknown state, %s" % block) File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line 574, in __exception_handler raise exception_class(message) InterfaceError: Unknown state, &2 65541 -1 !ERROR: COMMIT: transaction is aborted because of concurency conflicts, will ROLLBACK instead
I'm not sure what the problem is, but it exists in such a way that all operations that load or update information in MonetDB have to be put into a loop to continuously retry.
I'm using autocommit, but I would expect MonetDB to handle concurrency issues in a decent way, like maybe blocking a query until the other query finished. I'm not clobbering data, most of these are just bulk load operations that insert new rows.
Is there any way to address this issue without having to putting retry loops around all of the code I'm writing to write to the database?
73, Matthew W. Jones (KI4ZIB) http://matburt.net
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july
MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
There's no relational data in these tables, I guess I just need something internal to manage concurrency... like blocking table locks so I don't have to wait for a failure in order to manage concurrency on my own.
I noticed in one of my operations which was a single update to a table that no other processes were using that it gave me that error, informed me of the rollback, but the changes still took effect.... which is alarming since I will now need to:
1) Try the operation 2) If it fails, check to see if the change actually took effect 3) if the changes didn't take effect re-try the operation
This is disconcerting.
73, Matthew W. Jones (KI4ZIB) http://matburt.net
On Wed, Sep 2, 2009 at 3:02 PM, Martin Kersten Martin.Kersten@cwi.nlwrote:
Dear Matthew,
A quick initial reaction. MonetDB uses an optimistic concurrency control scheme, which is geared at high-read, low-write applications. If you hit the system with multiple concurrent writes on the same table, then indeed to secure acid properties, transactions are bound to fail. MonetDB indeed conforms to the semantics of a transaction safe system.
Blocking queries is equivalent to relation level locking. This is not the preferred way to go for the kernel, because someone will immediately ask for more granularities and als relaxation of the ACID principles.
There are several application scenarios to deal with this issue, but all depend on the semantics required.
regards, Martin
Matthew Jones wrote:
I have multiple processes reading and writing to the same table and I've noticed really persistent concurrency problems:
File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line 208, in execute self.__store_result(block) File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line 520, in __store_result self.__exception_handler(InterfaceError, "Unknown state, %s" % block) File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line 574, in __exception_handler raise exception_class(message) InterfaceError: Unknown state, &2 65541 -1 !ERROR: COMMIT: transaction is aborted because of concurency conflicts, will ROLLBACK instead
I'm not sure what the problem is, but it exists in such a way that all operations that load or update information in MonetDB have to be put into a loop to continuously retry.
I'm using autocommit, but I would expect MonetDB to handle concurrency issues in a decent way, like maybe blocking a query until the other query finished. I'm not clobbering data, most of these are just bulk load operations that insert new rows.
Is there any way to address this issue without having to putting retry loops around all of the code I'm writing to write to the database?
73, Matthew W. Jones (KI4ZIB) http://matburt.net
Let Crystal Reports handle the reporting - Free Crystal Reports 2008
30-Day
trial. Simplify your report design, integration and deployment - and
focus on
what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july
MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Curious, since I experienced this on a table that no one else was reading or writing to at the time. How does MonetDB determine that a concurrency violation has occurred? Does it look at table or record level read/write timestamps?
73, Matthew W. Jones (KI4ZIB) http://matburt.net
On Wed, Sep 2, 2009 at 3:33 PM, Matthew Jones mat@matburt.net wrote:
There's no relational data in these tables, I guess I just need something internal to manage concurrency... like blocking table locks so I don't have to wait for a failure in order to manage concurrency on my own.
I noticed in one of my operations which was a single update to a table that no other processes were using that it gave me that error, informed me of the rollback, but the changes still took effect.... which is alarming since I will now need to:
- Try the operation
- If it fails, check to see if the change actually took effect
- if the changes didn't take effect re-try the operation
This is disconcerting.
73, Matthew W. Jones (KI4ZIB) http://matburt.net
On Wed, Sep 2, 2009 at 3:02 PM, Martin Kersten Martin.Kersten@cwi.nlwrote:
Dear Matthew,
A quick initial reaction. MonetDB uses an optimistic concurrency control scheme, which is geared at high-read, low-write applications. If you hit the system with multiple concurrent writes on the same table, then indeed to secure acid properties, transactions are bound to fail. MonetDB indeed conforms to the semantics of a transaction safe system.
Blocking queries is equivalent to relation level locking. This is not the preferred way to go for the kernel, because someone will immediately ask for more granularities and als relaxation of the ACID principles.
There are several application scenarios to deal with this issue, but all depend on the semantics required.
regards, Martin
Matthew Jones wrote:
I have multiple processes reading and writing to the same table and I've noticed really persistent concurrency problems:
File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line 208, in execute self.__store_result(block) File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line 520, in __store_result self.__exception_handler(InterfaceError, "Unknown state, %s" %
block)
File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line 574, in __exception_handler raise exception_class(message) InterfaceError: Unknown state, &2 65541 -1 !ERROR: COMMIT: transaction is aborted because of concurency conflicts, will ROLLBACK instead
I'm not sure what the problem is, but it exists in such a way that all operations that load or update information in MonetDB have to be put into a loop to continuously retry.
I'm using autocommit, but I would expect MonetDB to handle concurrency issues in a decent way, like maybe blocking a query until the other query finished. I'm not clobbering data, most of these are just bulk load operations that insert new rows.
Is there any way to address this issue without having to putting retry loops around all of the code I'm writing to write to the database?
73, Matthew W. Jones (KI4ZIB) http://matburt.net
Let Crystal Reports handle the reporting - Free Crystal Reports 2008
30-Day
trial. Simplify your report design, integration and deployment - and
focus on
what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july
MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Matthew Jones wrote:
There's no relational data in these tables, I guess I just need something internal to manage concurrency... like blocking table locks so I don't have to wait for a failure in order to manage concurrency on my own.
you either have to serialize the updates/access at your client interface, or hook into the MAL kernel to get access to the general lock/semaphore module.
I noticed in one of my operations which was a single update to a table that no other processes were using that it gave me that error, informed
Optimistic concurrency control should also ensure that non of the concurrent transactions reads stale data. This becomes tricky if you have multiple user sessions with both autocommit and compound transactions, because then the timespans to trigger a transaction failure are larger.
me of the rollback, but the changes still took effect.... which is alarming since I will now need to:
- Try the operation
- If it fails, check to see if the change actually took effect
- if the changes didn't take effect re-try the operation
This is disconcerting.
73, Matthew W. Jones (KI4ZIB) http://matburt.net
On Wed, Sep 2, 2009 at 3:02 PM, Martin Kersten <Martin.Kersten@cwi.nl mailto:Martin.Kersten@cwi.nl> wrote:
Dear Matthew, A quick initial reaction. MonetDB uses an optimistic concurrency control scheme, which is geared at high-read, low-write applications. If you hit the system with multiple concurrent writes on the same table, then indeed to secure acid properties, transactions are bound to fail. MonetDB indeed conforms to the semantics of a transaction safe system. Blocking queries is equivalent to relation level locking. This is not the preferred way to go for the kernel, because someone will immediately ask for more granularities and als relaxation of the ACID principles. There are several application scenarios to deal with this issue, but all depend on the semantics required. regards, Martin Matthew Jones wrote: > I have multiple processes reading and writing to the same table and I've > noticed really persistent concurrency problems: > > File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line > 208, in execute > self.__store_result(block) > File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line > 520, in __store_result > self.__exception_handler(InterfaceError, "Unknown state, %s" % block) > File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line > 574, in __exception_handler > raise exception_class(message) > InterfaceError: Unknown state, &2 65541 -1 > !ERROR: COMMIT: transaction is aborted because of concurency conflicts, > will ROLLBACK instead > > I'm not sure what the problem is, but it exists in such a way that all > operations that load or update information in MonetDB have to be put > into a loop to continuously retry. > > I'm using autocommit, but I would expect MonetDB to handle concurrency > issues in a decent way, like maybe blocking a query until the other > query finished. I'm not clobbering data, most of these are just bulk > load operations that insert new rows. > > Is there any way to address this issue without having to putting retry > loops around all of the code I'm writing to write to the database? > > 73, > Matthew W. Jones (KI4ZIB) > http://matburt.net > > > ------------------------------------------------------------------------ > > ------------------------------------------------------------------------------ > Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day > trial. Simplify your report design, integration and deployment - and focus on > what you do best, core application coding. Discover what's new with > Crystal Reports now. http://p.sf.net/sfu/bobj-july > > > ------------------------------------------------------------------------ > > _______________________________________________ > MonetDB-users mailing list > MonetDB-users@lists.sourceforge.net <mailto:MonetDB-users@lists.sourceforge.net> > https://lists.sourceforge.net/lists/listinfo/monetdb-users ------------------------------------------------------------------------------ Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net <mailto:MonetDB-users@lists.sourceforge.net> https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july
MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Are you saying that if someone reads data from a table while another transaction is bulk-loading data that that will cause a concurrency conflict and a forced rollback?
I'm looking at the behavior of one of my long running batch loads and that looks to be the case... if so, that seems pretty extreme. I could understand protecting from simultaneous modification but it seems wrong to cause a write to fail for any read on a table.
73, Matthew W. Jones (KI4ZIB) http://matburt.net
On Wed, Sep 2, 2009 at 3:41 PM, Martin Kersten Martin.Kersten@cwi.nlwrote:
Matthew Jones wrote:
There's no relational data in these tables, I guess I just need something internal to manage concurrency... like blocking table locks so I don't have to wait for a failure in order to manage concurrency on my
own. you either have to serialize the updates/access at your client interface, or hook into the MAL kernel to get access to the general lock/semaphore module.
I noticed in one of my operations which was a single update to a table that no other processes were using that it gave me that error, informed
Optimistic concurrency control should also ensure that non of the concurrent transactions reads stale data. This becomes tricky if you have multiple user sessions with both autocommit and compound transactions, because then the timespans to trigger a transaction failure are larger.
me of the rollback, but the changes still took effect.... which is alarming since I will now need to:
- Try the operation
- If it fails, check to see if the change actually took effect
- if the changes didn't take effect re-try the operation
This is disconcerting.
73, Matthew W. Jones (KI4ZIB) http://matburt.net
On Wed, Sep 2, 2009 at 3:02 PM, Martin Kersten <Martin.Kersten@cwi.nl mailto:Martin.Kersten@cwi.nl> wrote:
Dear Matthew, A quick initial reaction. MonetDB uses an optimistic concurrencycontrol
scheme, which is geared at high-read, low-write applications. If youhit
the system with multiple concurrent writes on the same table, then indeed to secure acid properties, transactions are bound to fail. MonetDB indeed conforms to the semantics of a transaction safe system. Blocking queries is equivalent to relation level locking. This is not the preferred way to go for the kernel, because someone will immediately ask for more granularities and als relaxation of the ACID principles. There are several application scenarios to deal with this issue, but all depend on the semantics required. regards, Martin Matthew Jones wrote: > I have multiple processes reading and writing to the same table and I've > noticed really persistent concurrency problems: > > File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line > 208, in execute > self.__store_result(block) > File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line > 520, in __store_result > self.__exception_handler(InterfaceError, "Unknown state, %s" % block) > File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line > 574, in __exception_handler > raise exception_class(message) > InterfaceError: Unknown state, &2 65541 -1 > !ERROR: COMMIT: transaction is aborted because of concurency conflicts, > will ROLLBACK instead > > I'm not sure what the problem is, but it exists in such a way that all > operations that load or update information in MonetDB have to beput
> into a loop to continuously retry. > > I'm using autocommit, but I would expect MonetDB to handle concurrency > issues in a decent way, like maybe blocking a query until theother
> query finished. I'm not clobbering data, most of these are justbulk
> load operations that insert new rows. > > Is there any way to address this issue without having to putting retry > loops around all of the code I'm writing to write to the database? > > 73, > Matthew W. Jones (KI4ZIB) > http://matburt.net > > >
> >
> Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day > trial. Simplify your report design, integration and deployment - and focus on > what you do best, core application coding. Discover what's newwith
> Crystal Reports now. http://p.sf.net/sfu/bobj-july > > >
> > _______________________________________________ > MonetDB-users mailing list > MonetDB-users@lists.sourceforge.net <mailto:MonetDB-users@lists.sourceforge.net> > https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net <mailto:MonetDB-users@lists.sourceforge.net> https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008
30-Day
trial. Simplify your report design, integration and deployment - and
focus on
what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july
MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
On Wed, Sep 02, 2009 at 04:03:17PM -0400, Matthew Jones wrote:
Are you saying that if someone reads data from a table while another transaction is bulk-loading data that that will cause a concurrency conflict and a forced rollback?
I'm looking at the behavior of one of my long running batch loads and that looks to be the case... if so, that seems pretty extreme. I could understand protecting from simultaneous modification but it seems wrong to cause a write to fail for any read on a table.
writes conflict with reads, and reads with writes, just as writes conflict with writes; only multiple concurrent reads do not conlict with each other.
Stefan
73, Matthew W. Jones (KI4ZIB) http://matburt.net
On Wed, Sep 2, 2009 at 3:41 PM, Martin Kersten Martin.Kersten@cwi.nlwrote:
Matthew Jones wrote:
There's no relational data in these tables, I guess I just need something internal to manage concurrency... like blocking table locks so I don't have to wait for a failure in order to manage concurrency on my
own. you either have to serialize the updates/access at your client interface, or hook into the MAL kernel to get access to the general lock/semaphore module.
I noticed in one of my operations which was a single update to a table that no other processes were using that it gave me that error, informed
Optimistic concurrency control should also ensure that non of the concurrent transactions reads stale data. This becomes tricky if you have multiple user sessions with both autocommit and compound transactions, because then the timespans to trigger a transaction failure are larger.
me of the rollback, but the changes still took effect.... which is alarming since I will now need to:
- Try the operation
- If it fails, check to see if the change actually took effect
- if the changes didn't take effect re-try the operation
This is disconcerting.
73, Matthew W. Jones (KI4ZIB) http://matburt.net
On Wed, Sep 2, 2009 at 3:02 PM, Martin Kersten <Martin.Kersten@cwi.nl mailto:Martin.Kersten@cwi.nl> wrote:
Dear Matthew, A quick initial reaction. MonetDB uses an optimistic concurrencycontrol
scheme, which is geared at high-read, low-write applications. If youhit
the system with multiple concurrent writes on the same table, then indeed to secure acid properties, transactions are bound to fail. MonetDB indeed conforms to the semantics of a transaction safe system. Blocking queries is equivalent to relation level locking. This is not the preferred way to go for the kernel, because someone will immediately ask for more granularities and als relaxation of the ACID principles. There are several application scenarios to deal with this issue, but all depend on the semantics required. regards, Martin Matthew Jones wrote: > I have multiple processes reading and writing to the same table and I've > noticed really persistent concurrency problems: > > File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line > 208, in execute > self.__store_result(block) > File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line > 520, in __store_result > self.__exception_handler(InterfaceError, "Unknown state, %s" % block) > File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line > 574, in __exception_handler > raise exception_class(message) > InterfaceError: Unknown state, &2 65541 -1 > !ERROR: COMMIT: transaction is aborted because of concurency conflicts, > will ROLLBACK instead > > I'm not sure what the problem is, but it exists in such a way that all > operations that load or update information in MonetDB have to beput
> into a loop to continuously retry. > > I'm using autocommit, but I would expect MonetDB to handle concurrency > issues in a decent way, like maybe blocking a query until theother
> query finished. I'm not clobbering data, most of these are justbulk
> load operations that insert new rows. > > Is there any way to address this issue without having to putting retry > loops around all of the code I'm writing to write to the database? > > 73, > Matthew W. Jones (KI4ZIB) > http://matburt.net > > >
> >
> Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day > trial. Simplify your report design, integration and deployment - and focus on > what you do best, core application coding. Discover what's newwith
> Crystal Reports now. http://p.sf.net/sfu/bobj-july > > >
> > _______________________________________________ > MonetDB-users mailing list > MonetDB-users@lists.sourceforge.net <mailto:MonetDB-users@lists.sourceforge.net> > https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net <mailto:MonetDB-users@lists.sourceforge.net> https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008
30-Day
trial. Simplify your report design, integration and deployment - and
focus on
what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july
MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
If the write is Isolated from a read then should a read really be allowed to cause a concurrency failure on the write? The read only needs to be guaranteed to not see the table in a partial/incomplete state and the write's changes are isolated until it commits. A read conflicting with a write seems like something altogether different than a write conflicting with another write.
Am I missing something here?
73, Matthew W. Jones (KI4ZIB) http://matburt.net
On Wed, Sep 2, 2009 at 4:16 PM, Stefan Manegold Stefan.Manegold@cwi.nlwrote:
On Wed, Sep 02, 2009 at 04:03:17PM -0400, Matthew Jones wrote:
Are you saying that if someone reads data from a table while another transaction is bulk-loading data that that will cause a concurrency
conflict
and a forced rollback?
I'm looking at the behavior of one of my long running batch loads and
that
looks to be the case... if so, that seems pretty extreme. I could understand protecting from simultaneous modification but it seems wrong
to
cause a write to fail for any read on a table.
writes conflict with reads, and reads with writes, just as writes conflict with writes; only multiple concurrent reads do not conlict with each other.
Stefan
73, Matthew W. Jones (KI4ZIB) http://matburt.net
On Wed, Sep 2, 2009 at 3:41 PM, Martin Kersten <Martin.Kersten@cwi.nl wrote:
Matthew Jones wrote:
There's no relational data in these tables, I guess I just need something internal to manage concurrency... like blocking table locks
so
I don't have to wait for a failure in order to manage concurrency on
my
own. you either have to serialize the updates/access at your client
interface,
or hook into the MAL kernel to get access to the general lock/semaphore module.
I noticed in one of my operations which was a single update to a
table
that no other processes were using that it gave me that error,
informed
Optimistic concurrency control should also ensure that non of the concurrent transactions reads stale data. This becomes tricky if you have multiple user sessions with both
autocommit
and compound transactions, because then the timespans to trigger a transaction failure are larger.
me of the rollback, but the changes still took effect.... which is alarming since I will now need to:
- Try the operation
- If it fails, check to see if the change actually took effect
- if the changes didn't take effect re-try the operation
This is disconcerting.
73, Matthew W. Jones (KI4ZIB) http://matburt.net
On Wed, Sep 2, 2009 at 3:02 PM, Martin Kersten <
Martin.Kersten@cwi.nl
mailto:Martin.Kersten@cwi.nl> wrote:
Dear Matthew, A quick initial reaction. MonetDB uses an optimistic concurrencycontrol
scheme, which is geared at high-read, low-write applications. Ifyou
hit
the system with multiple concurrent writes on the same table,then
indeed to secure acid properties, transactions are bound to fail.MonetDB
indeed conforms to the semantics of a transaction safe system. Blocking queries is equivalent to relation level locking. This is not the preferred way to go for the kernel, becausesomeone
will immediately ask for more granularities and als relaxation of the ACID principles. There are several application scenarios to deal with this issue, but all depend on the semantics required. regards, Martin Matthew Jones wrote: > I have multiple processes reading and writing to the sametable
and I've > noticed really persistent concurrency problems: > > File"/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py",
line > 208, in execute > self.__store_result(block) > File"/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py",
line > 520, in __store_result > self.__exception_handler(InterfaceError, "Unknown state,%s"
% block) > File"/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py",
line > 574, in __exception_handler > raise exception_class(message) > InterfaceError: Unknown state, &2 65541 -1 > !ERROR: COMMIT: transaction is aborted because of concurency conflicts, > will ROLLBACK instead > > I'm not sure what the problem is, but it exists in such a way that all > operations that load or update information in MonetDB have tobe
put
> into a loop to continuously retry. > > I'm using autocommit, but I would expect MonetDB to handle concurrency > issues in a decent way, like maybe blocking a query until theother
> query finished. I'm not clobbering data, most of these arejust
bulk
> load operations that insert new rows. > > Is there any way to address this issue without having toputting
retry > loops around all of the code I'm writing to write to thedatabase?
> > 73, > Matthew W. Jones (KI4ZIB) > http://matburt.net > > >
> >
> Let Crystal Reports handle the reporting - Free CrystalReports
2008 30-Day > trial. Simplify your report design, integration and deployment
and focus on > what you do best, core application coding. Discover what's newwith
> Crystal Reports now. http://p.sf.net/sfu/bobj-july > > >
> > _______________________________________________ > MonetDB-users mailing list > MonetDB-users@lists.sourceforge.net <mailto:MonetDB-users@lists.sourceforge.net> > https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports2008
30-Day trial. Simplify your report design, integration and deployment -and
focus on what you do best, core application coding. Discover what's newwith
Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net <mailto:MonetDB-users@lists.sourceforge.net> https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008
30-Day
trial. Simplify your report design, integration and deployment - and
focus on
what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july
MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008
30-Day
trial. Simplify your report design, integration and deployment - and
focus
on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008
30-Day
trial. Simplify your report design, integration and deployment - and
focus on
what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
-- | Dr. Stefan Manegold | mailto:Stefan.Manegold@cwi.nl | | CWI, P.O.Box 94079 | http://www.cwi.nl/~manegold/http://www.cwi.nl/%7Emanegold/ | | 1090 GB Amsterdam | Tel.: +31 (20) 592-4212 | | The Netherlands | Fax : +31 (20) 592-4312 |
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Hi,
Snapshot isolation in principle only needs to abort write-write conflicts.
However, MonetDB aims at serializabilty. Maybe we should consider offering SI as well?
In addition to that, my advise has been been for some time to consider row-wise write-write conflict checking rather than full-table.
Peter
If the write is Isolated from a read then should a read really be allowed to cause a concurrency failure on the write? The read only needs to be guaranteed to not see the table in a partial/incomplete state and the write's changes are isolated until it commits. A read conflicting with a write seems like something altogether different than a write conflicting with another write.
Am I missing something here?
73, Matthew W. Jones (KI4ZIB) http://matburt.net
On Wed, Sep 2, 2009 at 4:16 PM, Stefan Manegold Stefan.Manegold@cwi.nlwrote:
On Wed, Sep 02, 2009 at 04:03:17PM -0400, Matthew Jones wrote:
Are you saying that if someone reads data from a table while another transaction is bulk-loading data that that will cause a concurrency
conflict
and a forced rollback?
I'm looking at the behavior of one of my long running batch loads and
that
looks to be the case... if so, that seems pretty extreme. I could understand protecting from simultaneous modification but it seems wrong
to
cause a write to fail for any read on a table.
writes conflict with reads, and reads with writes, just as writes conflict with writes; only multiple concurrent reads do not conlict with each other.
Stefan
73, Matthew W. Jones (KI4ZIB) http://matburt.net
On Wed, Sep 2, 2009 at 3:41 PM, Martin Kersten <Martin.Kersten@cwi.nl wrote:
Matthew Jones wrote:
There's no relational data in these tables, I guess I just need something internal to manage concurrency... like blocking table locks
so
I don't have to wait for a failure in order to manage concurrency on
my
own. you either have to serialize the updates/access at your client
interface,
or hook into the MAL kernel to get access to the general lock/semaphore module.
I noticed in one of my operations which was a single update to a
table
that no other processes were using that it gave me that error,
informed
Optimistic concurrency control should also ensure that non of the concurrent transactions reads stale data. This becomes tricky if you have multiple user sessions with both
autocommit
and compound transactions, because then the timespans to trigger a transaction failure are larger.
me of the rollback, but the changes still took effect.... which is alarming since I will now need to:
- Try the operation
- If it fails, check to see if the change actually took effect
- if the changes didn't take effect re-try the operation
This is disconcerting.
73, Matthew W. Jones (KI4ZIB) http://matburt.net
On Wed, Sep 2, 2009 at 3:02 PM, Martin Kersten <
Martin.Kersten@cwi.nl
mailto:Martin.Kersten@cwi.nl> wrote:
Dear Matthew, A quick initial reaction. MonetDB uses an optimistic concurrencycontrol
scheme, which is geared at high-read, low-write applications. Ifyou
hit
the system with multiple concurrent writes on the same table,then
indeed to secure acid properties, transactions are bound to fail.MonetDB
indeed conforms to the semantics of a transaction safe system. Blocking queries is equivalent to relation level locking. This is not the preferred way to go for the kernel, becausesomeone
will immediately ask for more granularities and als relaxation of the ACID principles. There are several application scenarios to deal with this issue, but all depend on the semantics required. regards, Martin Matthew Jones wrote: > I have multiple processes reading and writing to the sametable
and I've > noticed really persistent concurrency problems: > > File"/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py",
line > 208, in execute > self.__store_result(block) > File"/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py",
line > 520, in __store_result > self.__exception_handler(InterfaceError, "Unknown state,%s"
% block) > File"/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py",
line > 574, in __exception_handler > raise exception_class(message) > InterfaceError: Unknown state, &2 65541 -1 > !ERROR: COMMIT: transaction is aborted because of concurency conflicts, > will ROLLBACK instead > > I'm not sure what the problem is, but it exists in such a way that all > operations that load or update information in MonetDB have tobe
put
> into a loop to continuously retry. > > I'm using autocommit, but I would expect MonetDB to handle concurrency > issues in a decent way, like maybe blocking a query until theother
> query finished. I'm not clobbering data, most of these arejust
bulk
> load operations that insert new rows. > > Is there any way to address this issue without having toputting
retry > loops around all of the code I'm writing to write to thedatabase?
> > 73, > Matthew W. Jones (KI4ZIB) > http://matburt.net > > >
> >
> Let Crystal Reports handle the reporting - Free CrystalReports
2008 30-Day > trial. Simplify your report design, integration and deployment
and focus on > what you do best, core application coding. Discover what's newwith
> Crystal Reports now. http://p.sf.net/sfu/bobj-july > > >
> > _______________________________________________ > MonetDB-users mailing list > MonetDB-users@lists.sourceforge.net <mailto:MonetDB-users@lists.sourceforge.net> > https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports2008
30-Day trial. Simplify your report design, integration and deployment -and
focus on what you do best, core application coding. Discover what's newwith
Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net <mailto:MonetDB-users@lists.sourceforge.net> https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008
30-Day
trial. Simplify your report design, integration and deployment - and
focus on
what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july
MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008
30-Day
trial. Simplify your report design, integration and deployment - and
focus
on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008
30-Day
trial. Simplify your report design, integration and deployment - and
focus on
what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
-- | Dr. Stefan Manegold | mailto:Stefan.Manegold@cwi.nl | | CWI, P.O.Box 94079 | http://www.cwi.nl/~manegold/http://www.cwi.nl/%7Emanegold/ | | 1090 GB Amsterdam | Tel.: +31 (20) 592-4212 | | The Netherlands | Fax : +31 (20) 592-4312 |
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july_______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
On Wed, Sep 02, 2009 at 11:06:12PM +0200, Peter Boncz wrote:
Hi,
Snapshot isolation in principle only needs to abort write-write conflicts.
However, MonetDB aims at serializabilty. Maybe we should consider offering SI
Yes MonetDB has serializability. Still a single only read transaction shouldn't affect a writing transaction.
Niels
as well?
In addition to that, my advise has been been for some time to consider row-wise write-write conflict checking rather than full-table.
Peter
If the write is Isolated from a read then should a read really be allowed to cause a concurrency failure on the write? The read only needs to be guaranteed to not see the table in a partial/incomplete state and the write's changes are isolated until it commits. A read conflicting with a write seems like something altogether different than a write conflicting with another write.
Am I missing something here?
73, Matthew W. Jones (KI4ZIB) http://matburt.net
On Wed, Sep 2, 2009 at 4:16 PM, Stefan Manegold Stefan.Manegold@cwi.nlwrote:
On Wed, Sep 02, 2009 at 04:03:17PM -0400, Matthew Jones wrote:
Are you saying that if someone reads data from a table while another transaction is bulk-loading data that that will cause a concurrency
conflict
and a forced rollback?
I'm looking at the behavior of one of my long running batch loads and
that
looks to be the case... if so, that seems pretty extreme. I could understand protecting from simultaneous modification but it seems wrong
to
cause a write to fail for any read on a table.
writes conflict with reads, and reads with writes, just as writes conflict with writes; only multiple concurrent reads do not conlict with each other.
Stefan
73, Matthew W. Jones (KI4ZIB) http://matburt.net
On Wed, Sep 2, 2009 at 3:41 PM, Martin Kersten <Martin.Kersten@cwi.nl wrote:
Matthew Jones wrote:
There's no relational data in these tables, I guess I just need something internal to manage concurrency... like blocking table locks
so
I don't have to wait for a failure in order to manage concurrency on
my
own. you either have to serialize the updates/access at your client
interface,
or hook into the MAL kernel to get access to the general lock/semaphore module.
I noticed in one of my operations which was a single update to a
table
that no other processes were using that it gave me that error,
informed
Optimistic concurrency control should also ensure that non of the concurrent transactions reads stale data. This becomes tricky if you have multiple user sessions with both
autocommit
and compound transactions, because then the timespans to trigger a transaction failure are larger.
me of the rollback, but the changes still took effect.... which is alarming since I will now need to:
- Try the operation
- If it fails, check to see if the change actually took effect
- if the changes didn't take effect re-try the operation
This is disconcerting.
73, Matthew W. Jones (KI4ZIB) http://matburt.net
On Wed, Sep 2, 2009 at 3:02 PM, Martin Kersten <
Martin.Kersten@cwi.nl
mailto:Martin.Kersten@cwi.nl> wrote:
Dear Matthew, A quick initial reaction. MonetDB uses an optimistic concurrencycontrol
scheme, which is geared at high-read, low-write applications. Ifyou
hit
the system with multiple concurrent writes on the same table,then
indeed to secure acid properties, transactions are bound to fail.MonetDB
indeed conforms to the semantics of a transaction safe system. Blocking queries is equivalent to relation level locking. This is not the preferred way to go for the kernel, becausesomeone
will immediately ask for more granularities and als relaxation of the ACID principles. There are several application scenarios to deal with this issue, but all depend on the semantics required. regards, Martin Matthew Jones wrote: > I have multiple processes reading and writing to the sametable
and I've > noticed really persistent concurrency problems: > > File"/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py",
line > 208, in execute > self.__store_result(block) > File"/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py",
line > 520, in __store_result > self.__exception_handler(InterfaceError, "Unknown state,%s"
% block) > File"/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py",
line > 574, in __exception_handler > raise exception_class(message) > InterfaceError: Unknown state, &2 65541 -1 > !ERROR: COMMIT: transaction is aborted because of concurency conflicts, > will ROLLBACK instead > > I'm not sure what the problem is, but it exists in such a way that all > operations that load or update information in MonetDB have tobe
put
> into a loop to continuously retry. > > I'm using autocommit, but I would expect MonetDB to handle concurrency > issues in a decent way, like maybe blocking a query until theother
> query finished. I'm not clobbering data, most of these arejust
bulk
> load operations that insert new rows. > > Is there any way to address this issue without having toputting
retry > loops around all of the code I'm writing to write to thedatabase?
> > 73, > Matthew W. Jones (KI4ZIB) > http://matburt.net > > >
> >
> Let Crystal Reports handle the reporting - Free CrystalReports
2008 30-Day > trial. Simplify your report design, integration and deployment
and focus on > what you do best, core application coding. Discover what's newwith
> Crystal Reports now. http://p.sf.net/sfu/bobj-july > > >
> > _______________________________________________ > MonetDB-users mailing list > MonetDB-users@lists.sourceforge.net <mailto:MonetDB-users@lists.sourceforge.net> > https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports2008
30-Day trial. Simplify your report design, integration and deployment -and
focus on what you do best, core application coding. Discover what's newwith
Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net <mailto:MonetDB-users@lists.sourceforge.net> https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008
30-Day
trial. Simplify your report design, integration and deployment - and
focus on
what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july
MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008
30-Day
trial. Simplify your report design, integration and deployment - and
focus
on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008
30-Day
trial. Simplify your report design, integration and deployment - and
focus on
what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
-- | Dr. Stefan Manegold | mailto:Stefan.Manegold@cwi.nl | | CWI, P.O.Box 94079 | http://www.cwi.nl/~manegold/http://www.cwi.nl/%7Emanegold/ | | 1090 GB Amsterdam | Tel.: +31 (20) 592-4212 | | The Netherlands | Fax : +31 (20) 592-4312 |
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july_______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
Matthew Jones wrote:
I have multiple processes reading and writing to the same table and I've noticed really persistent concurrency problems:
File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line 208, in execute self.__store_result(block) File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line 520, in __store_result self.__exception_handler(InterfaceError, "Unknown state, %s" % block) File "/usr/lib/python2.5/site-packages/monetdb/sql/cursors.py", line 574, in __exception_handler raise exception_class(message) InterfaceError: Unknown state, &2 65541 -1 !ERROR: COMMIT: transaction is aborted because of concurency conflicts, will ROLLBACK instead
hi all,
This is a bug in the python API also, it doesn't handle the error properly. The desired behavior would be that a DatabaseError exception is raised. Will fix asap.
- gijs
I'm not sure what the problem is, but it exists in such a way that all operations that load or update information in MonetDB have to be put into a loop to continuously retry.
I'm using autocommit, but I would expect MonetDB to handle concurrency issues in a decent way, like maybe blocking a query until the other query finished. I'm not clobbering data, most of these are just bulk load operations that insert new rows.
Is there any way to address this issue without having to putting retry loops around all of the code I'm writing to write to the database?
73, Matthew W. Jones (KI4ZIB) http://matburt.net
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july
MonetDB-users mailing list MonetDB-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-users
participants (6)
-
Gijs Molenaar -
Martin Kersten -
Matthew Jones -
Niels Nes -
Peter Boncz -
Stefan Manegold