You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by "C. Michael Pilato" <cm...@collab.net> on 2008/11/14 19:41:22 UTC
FSFS safety violation caught by testing? (r34197)
cmpilato@tigris.org wrote:
> Author: cmpilato
> Date: Fri Nov 14 11:34:25 2008
> New Revision: 34197
>
> Log:
> Bang support for --server-minor-version testing into the C test
> framework, and use it in the FS and Repos tests.
>
> NOTE: Only one failure that I didn't expect turned up:
>
> $ ./fs-test --server-minor-version=4 --fs-type=fsfs
> ...
> subversion/tests/libsvn_fs/fs-test.c:369: (apr_err=160000)
> svn_tests: the transaction name '0-3' was reused
> FAIL: lt-fs-test 5: check that transaction names are not reused
> ...
Does an FSFS hacker want to comment on this? Do we have a problem when
using trunk code on --pre-1.5-compatible FSFS repositories?
--
C. Michael Pilato <cm...@collab.net>
CollabNet <> www.collab.net <> Distributed Development On Demand
Re: FSFS safety violation caught by testing? (r34197)
Posted by Blair Zajac <bl...@orcaware.com>.
C. Michael Pilato wrote:
> Blair Zajac wrote:
>> On Nov 14, 2008, at 7:53 PM, C. Michael Pilato wrote:
>>
>>> C. Michael Pilato wrote:
>>>> C. Michael Pilato wrote:
>>>>> David Glasser wrote:
>>>>>> Anyway, I think the transaction name algorithm changed in 1.5; check
>>>>>> libsvn_fs_fs/structure for details.
>>>> Hrm. Interestingly, the test doesn't care about naming schemes. It
>>>> creates
>>>> a bunch of transactions, remembers their names, and aborts them all.
>>>> The it
>>>> does it again to see if any of the transaction names got reused.
>>>> I'll look
>>>> into this a little bit more closely.
>>> Well, we have a decision to make. If we think it's a bug that
>>> transaction
>>> names can be reused, then FSFS repositories created with Subversion
>>> 1.4 have
>>> this bug. I don't understand that ramifications of txn reuse (which
>>> is why
>>> I've Cc: Blair -- I remember this topic coming up on the list and being
>>> driven by him).
>>>
>>> If we'd rather ignore the problem, it's easy enough to make the code bail
>>> out when testing FSFS with 1.4-era repository versions. I'm happy to
>>> make
>>> that change -- just let me know.
>> I put this feature in to support my svn_fs.h and svn_repos.h RPC API. I
>> found that when you had a client begin a transaction and another one use
>> it, then the first one close it and begin another one, the second client
>> would not know that it was not using the original any more (it should
>> have received a txn does not exist error instead).
>>
>> I don't need this feature in 1.4, so having txn ids be non-unique in 1.4
>> is fine with me.
>
> So, you don't need the "feature" of having processes not accidentally
> reusing old transaction handles in 1.4.
That's right. Our svn backend is using 1.5/fsfs so our system is fine.
> Is that kind of process isolation a
> feature? It sounds to me like transactions that can be reused unaware is a
> recipe for having bogus commits get made, is it not? That sounds like a bug
> to me. But I admit to not fully understanding the interactions here. Can
> you fill me in?
Well, I don't believe it's an issue with the svn client, svnserve and
mod_dav_svn because only a single client ever works with a transaction. When
the transaction is committed or rolled back, there is no other client that is
currently running that is using that txn id, so there's no chance of misuse.
There's no program that I'm aware of that uses a txn by two or more
process/threads where the another process/thread isn't aware of the transaction
committing or rolling back.
Blair
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: FSFS safety violation caught by testing? (r34197)
Posted by "C. Michael Pilato" <cm...@collab.net>.
David Glasser wrote:
> I'm pretty sure we only decided to declare "transaction IDs are never
> reused" in 1.5. (If we even declared it then.)
Okey dokey. I've made the test bail out early for FSFS with pre-1.5
repositories.
--
C. Michael Pilato <cm...@collab.net>
CollabNet <> www.collab.net <> Distributed Development On Demand
Re: FSFS safety violation caught by testing? (r34197)
Posted by David Glasser <gl...@davidglasser.net>.
I'm pretty sure we only decided to declare "transaction IDs are never
reused" in 1.5. (If we even declared it then.)
--dave
On Sat, Nov 15, 2008 at 4:41 PM, C. Michael Pilato <cm...@collab.net> wrote:
> Blair Zajac wrote:
>>
>> On Nov 14, 2008, at 7:53 PM, C. Michael Pilato wrote:
>>
>>> C. Michael Pilato wrote:
>>>> C. Michael Pilato wrote:
>>>>> David Glasser wrote:
>>>>>> Anyway, I think the transaction name algorithm changed in 1.5; check
>>>>>> libsvn_fs_fs/structure for details.
>>>>
>>>> Hrm. Interestingly, the test doesn't care about naming schemes. It
>>>> creates
>>>> a bunch of transactions, remembers their names, and aborts them all.
>>>> The it
>>>> does it again to see if any of the transaction names got reused.
>>>> I'll look
>>>> into this a little bit more closely.
>>>
>>> Well, we have a decision to make. If we think it's a bug that
>>> transaction
>>> names can be reused, then FSFS repositories created with Subversion
>>> 1.4 have
>>> this bug. I don't understand that ramifications of txn reuse (which
>>> is why
>>> I've Cc: Blair -- I remember this topic coming up on the list and being
>>> driven by him).
>>>
>>> If we'd rather ignore the problem, it's easy enough to make the code bail
>>> out when testing FSFS with 1.4-era repository versions. I'm happy to
>>> make
>>> that change -- just let me know.
>>
>> I put this feature in to support my svn_fs.h and svn_repos.h RPC API. I
>> found that when you had a client begin a transaction and another one use
>> it, then the first one close it and begin another one, the second client
>> would not know that it was not using the original any more (it should
>> have received a txn does not exist error instead).
>>
>> I don't need this feature in 1.4, so having txn ids be non-unique in 1.4
>> is fine with me.
>
> So, you don't need the "feature" of having processes not accidentally
> reusing old transaction handles in 1.4. Is that kind of process isolation a
> feature? It sounds to me like transactions that can be reused unaware is a
> recipe for having bogus commits get made, is it not? That sounds like a bug
> to me. But I admit to not fully understanding the interactions here. Can
> you fill me in?
>
> --
> C. Michael Pilato <cm...@collab.net>
> CollabNet <> www.collab.net <> Distributed Development On Demand
>
>
--
David Glasser | glasser@davidglasser.net | http://www.davidglasser.net/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: FSFS safety violation caught by testing? (r34197)
Posted by "C. Michael Pilato" <cm...@collab.net>.
Blair Zajac wrote:
>
> On Nov 14, 2008, at 7:53 PM, C. Michael Pilato wrote:
>
>> C. Michael Pilato wrote:
>>> C. Michael Pilato wrote:
>>>> David Glasser wrote:
>>>>> Anyway, I think the transaction name algorithm changed in 1.5; check
>>>>> libsvn_fs_fs/structure for details.
>>>
>>> Hrm. Interestingly, the test doesn't care about naming schemes. It
>>> creates
>>> a bunch of transactions, remembers their names, and aborts them all.
>>> The it
>>> does it again to see if any of the transaction names got reused.
>>> I'll look
>>> into this a little bit more closely.
>>
>> Well, we have a decision to make. If we think it's a bug that
>> transaction
>> names can be reused, then FSFS repositories created with Subversion
>> 1.4 have
>> this bug. I don't understand that ramifications of txn reuse (which
>> is why
>> I've Cc: Blair -- I remember this topic coming up on the list and being
>> driven by him).
>>
>> If we'd rather ignore the problem, it's easy enough to make the code bail
>> out when testing FSFS with 1.4-era repository versions. I'm happy to
>> make
>> that change -- just let me know.
>
> I put this feature in to support my svn_fs.h and svn_repos.h RPC API. I
> found that when you had a client begin a transaction and another one use
> it, then the first one close it and begin another one, the second client
> would not know that it was not using the original any more (it should
> have received a txn does not exist error instead).
>
> I don't need this feature in 1.4, so having txn ids be non-unique in 1.4
> is fine with me.
So, you don't need the "feature" of having processes not accidentally
reusing old transaction handles in 1.4. Is that kind of process isolation a
feature? It sounds to me like transactions that can be reused unaware is a
recipe for having bogus commits get made, is it not? That sounds like a bug
to me. But I admit to not fully understanding the interactions here. Can
you fill me in?
--
C. Michael Pilato <cm...@collab.net>
CollabNet <> www.collab.net <> Distributed Development On Demand
Re: FSFS safety violation caught by testing? (r34197)
Posted by Blair Zajac <bl...@orcaware.com>.
On Nov 14, 2008, at 7:53 PM, C. Michael Pilato wrote:
> C. Michael Pilato wrote:
>> C. Michael Pilato wrote:
>>> David Glasser wrote:
>>>> Anyway, I think the transaction name algorithm changed in 1.5;
>>>> check
>>>> libsvn_fs_fs/structure for details.
>>
>> Hrm. Interestingly, the test doesn't care about naming schemes.
>> It creates
>> a bunch of transactions, remembers their names, and aborts them
>> all. The it
>> does it again to see if any of the transaction names got reused.
>> I'll look
>> into this a little bit more closely.
>
> Well, we have a decision to make. If we think it's a bug that
> transaction
> names can be reused, then FSFS repositories created with Subversion
> 1.4 have
> this bug. I don't understand that ramifications of txn reuse (which
> is why
> I've Cc: Blair -- I remember this topic coming up on the list and
> being
> driven by him).
>
> If we'd rather ignore the problem, it's easy enough to make the code
> bail
> out when testing FSFS with 1.4-era repository versions. I'm happy
> to make
> that change -- just let me know.
I put this feature in to support my svn_fs.h and svn_repos.h RPC API.
I found that when you had a client begin a transaction and another one
use it, then the first one close it and begin another one, the second
client would not know that it was not using the original any more (it
should have received a txn does not exist error instead).
I don't need this feature in 1.4, so having txn ids be non-unique in
1.4 is fine with me.
Regards,
Blair
--
Blair Zajac, Ph.D.
CTO, OrcaWare Technologies
<bl...@orcaware.com>
Subversion training, consulting and support
http://www.orcaware.com/svn/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: FSFS safety violation caught by testing? (r34197)
Posted by "C. Michael Pilato" <cm...@collab.net>.
C. Michael Pilato wrote:
> C. Michael Pilato wrote:
>> David Glasser wrote:
>>> Anyway, I think the transaction name algorithm changed in 1.5; check
>>> libsvn_fs_fs/structure for details.
>
> Hrm. Interestingly, the test doesn't care about naming schemes. It creates
> a bunch of transactions, remembers their names, and aborts them all. The it
> does it again to see if any of the transaction names got reused. I'll look
> into this a little bit more closely.
Well, we have a decision to make. If we think it's a bug that transaction
names can be reused, then FSFS repositories created with Subversion 1.4 have
this bug. I don't understand that ramifications of txn reuse (which is why
I've Cc: Blair -- I remember this topic coming up on the list and being
driven by him).
If we'd rather ignore the problem, it's easy enough to make the code bail
out when testing FSFS with 1.4-era repository versions. I'm happy to make
that change -- just let me know.
--
C. Michael Pilato <cm...@collab.net>
CollabNet <> www.collab.net <> Distributed Development On Demand
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: FSFS safety violation caught by testing? (r34197)
Posted by "C. Michael Pilato" <cm...@collab.net>.
C. Michael Pilato wrote:
> David Glasser wrote:
>> Anyway, I think the transaction name algorithm changed in 1.5; check
>> libsvn_fs_fs/structure for details.
Hrm. Interestingly, the test doesn't care about naming schemes. It creates
a bunch of transactions, remembers their names, and aborts them all. The it
does it again to see if any of the transaction names got reused. I'll look
into this a little bit more closely.
--
C. Michael Pilato <cm...@collab.net>
CollabNet <> www.collab.net <> Distributed Development On Demand
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: FSFS safety violation caught by testing? (r34197)
Posted by "C. Michael Pilato" <cm...@collab.net>.
David Glasser wrote:
> On Fri, Nov 14, 2008 at 3:25 PM, David Glasser <gl...@davidglasser.net> wrote:
>> On Fri, Nov 14, 2008 at 11:41 AM, C. Michael Pilato <cm...@collab.net> wrote:
>>> cmpilato@tigris.org wrote:
>>>> Author: cmpilato
>>>> Date: Fri Nov 14 11:34:25 2008
>>>> New Revision: 34197
>>>>
>>>> Log:
>>>> Bang support for --server-minor-version testing into the C test
>>>> framework, and use it in the FS and Repos tests.
>>>>
>>>> NOTE: Only one failure that I didn't expect turned up:
>>>>
>>>> $ ./fs-test --server-minor-version=4 --fs-type=fsfs
>>>> ...
>>>> subversion/tests/libsvn_fs/fs-test.c:369: (apr_err=160000)
>>>> svn_tests: the transaction name '0-3' was reused
>>>> FAIL: lt-fs-test 5: check that transaction names are not reused
>>>> ...
>>> Does an FSFS hacker want to comment on this? Do we have a problem when
>>> using trunk code on --pre-1.5-compatible FSFS repositories?
>> Unless something has changed the C tests don't pay attention to
>> --server-minor-version.
>
> ... OK, OK, now I read the log message :)
>
> Unconvinced that --server-minor-version makes sense for non-RA tests, but sure.
I don't see how the C tests would stand out as unique against, say, ra-local
tests in this manner. We need a way to make sure that newer FS code doesn't
croak on old formats (or *does* croak predictably as the case might be).
This seemed like the easiest way to get there from here.
> Anyway, I think the transaction name algorithm changed in 1.5; check
> libsvn_fs_fs/structure for details.
Okey dokey.
--
C. Michael Pilato <cm...@collab.net>
CollabNet <> www.collab.net <> Distributed Development On Demand
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: FSFS safety violation caught by testing? (r34197)
Posted by David Glasser <gl...@davidglasser.net>.
On Fri, Nov 14, 2008 at 3:25 PM, David Glasser <gl...@davidglasser.net> wrote:
> On Fri, Nov 14, 2008 at 11:41 AM, C. Michael Pilato <cm...@collab.net> wrote:
>> cmpilato@tigris.org wrote:
>>> Author: cmpilato
>>> Date: Fri Nov 14 11:34:25 2008
>>> New Revision: 34197
>>>
>>> Log:
>>> Bang support for --server-minor-version testing into the C test
>>> framework, and use it in the FS and Repos tests.
>>>
>>> NOTE: Only one failure that I didn't expect turned up:
>>>
>>> $ ./fs-test --server-minor-version=4 --fs-type=fsfs
>>> ...
>>> subversion/tests/libsvn_fs/fs-test.c:369: (apr_err=160000)
>>> svn_tests: the transaction name '0-3' was reused
>>> FAIL: lt-fs-test 5: check that transaction names are not reused
>>> ...
>>
>> Does an FSFS hacker want to comment on this? Do we have a problem when
>> using trunk code on --pre-1.5-compatible FSFS repositories?
>
> Unless something has changed the C tests don't pay attention to
> --server-minor-version.
... OK, OK, now I read the log message :)
Unconvinced that --server-minor-version makes sense for non-RA tests, but sure.
Anyway, I think the transaction name algorithm changed in 1.5; check
libsvn_fs_fs/structure for details.
--dave
--
David Glasser | glasser@davidglasser.net | http://www.davidglasser.net/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: FSFS safety violation caught by testing? (r34197)
Posted by David Glasser <gl...@davidglasser.net>.
On Fri, Nov 14, 2008 at 11:41 AM, C. Michael Pilato <cm...@collab.net> wrote:
> cmpilato@tigris.org wrote:
>> Author: cmpilato
>> Date: Fri Nov 14 11:34:25 2008
>> New Revision: 34197
>>
>> Log:
>> Bang support for --server-minor-version testing into the C test
>> framework, and use it in the FS and Repos tests.
>>
>> NOTE: Only one failure that I didn't expect turned up:
>>
>> $ ./fs-test --server-minor-version=4 --fs-type=fsfs
>> ...
>> subversion/tests/libsvn_fs/fs-test.c:369: (apr_err=160000)
>> svn_tests: the transaction name '0-3' was reused
>> FAIL: lt-fs-test 5: check that transaction names are not reused
>> ...
>
> Does an FSFS hacker want to comment on this? Do we have a problem when
> using trunk code on --pre-1.5-compatible FSFS repositories?
Unless something has changed the C tests don't pay attention to
--server-minor-version.
--dave
--
David Glasser | glasser@davidglasser.net | http://www.davidglasser.net/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org