Can anybody see that problem (Or do I misunderstand the manual page)?
> -Original Message-
> From: openais-boun...@lists.linux-foundation.org [mailto:openais-
> boun...@lists.linux-foundation.org] On Behalf Of Dietmar Maurer
> Sent: Donnerstag, 23. Juni 2011 07:5
I just noticed that cororsync is unable to set the RT scheduler priority when I
enable CONFIG_RT_GROUP_SCHED (return EPERM). Is that expected?
kernel: http://download.openvz.org/kernel/branches/rhel6-2.6.32/042stab018.1/
- Dietmar
___
Openais mailing
There are two versions of strerror_r ('man strerror_r')
corosync configure.ac. defined -D_GNU_SOURCE, so we use the GNU specific
version.
The manual page claims:
The GNU-specific strerror_r() returns a pointer to a string containing
the error message. This may be either a pointer
--- corosync-1.3.1/exec/main.c.org 2011-06-23 06:58:38.0 +0200
+++ corosync-1.3.1/exec/main.c 2011-06-23 06:58:44.0 +0200
@@ -1230,10 +1230,10 @@
if (res == -1) {
char error_str[100];
strerror_r (errno, error_str,
see http://www.corosync.org/git/?p=corosync.git;a=summary
I thought there should be a tag for v1.3.1?
And what is the difference between head 'flatiron' and 'flatiron-1.3'?
- Dietmar
___
Openais mailing list
Openais@lists.linux-foundation.org
https://
> I pload is not destrustive, but it does not play nicely. It is a tool to
> compare
> sending messages without IPC been involved (compare with cpgbench not to be
> used at the same time).
Many thanks for the explanation. I just wanted to do some stress testing, and
thought the pload service was
> > Any idea?
>
> Don´t run pload.. that´s the issue. It´s not meant to be executed on something
> that needs to survive.
Ok, thanks (I guess it would be a good idea to mention that in the manual page).
- Dietmar
___
Openais mailing list
Openais@lists
osync-pload
Any idea?
>
> Fabio
>
> On 12/22/2010 9:52 AM, Dietmar Maurer wrote:
> > Corosync v1.3.0 (single node)
> >
> > Debian Squeeze AMD64 with latest 2.6.32 kernel
> >
> >
> >
> > When I run "corosync-pload" it prints:
>
Corosync v1.3.0 (single node)
Debian Squeeze AMD64 with latest 2.6.32 kernel
When I run "corosync-pload" it prints:
# corosync-pload
Init result 1
The process never stops (but I can stop it with cntrl-c), but it seems to work
anyways:
Dec 22 09:32:46 maui corosync[2409]: [PLOAD ] 150 Wri
> > I thought the return value is used to indicate errors, but if I simply
> > return a negative number my client hang forever.
>
> Currently if you return anything but -ENOBUFS or -EAGAIN the server will
> remove the socket from the poll loop.
I can't reproduce that behavior. My server continues
Hi all,
I am playing around with libqb writing my first test server. Normal operation
work
quite good so far. I just wonder how to handle errors on the server side,
especially
in qb_ipcs_msg_process_fn:
typedef int32_t (*qb_ipcs_msg_process_fn) (qb_ipcs_connection_t *c,
void *da
I assume those newlines "\n" should not be there?
--- ipcs.c.org 2010-12-03 09:32:58.0 +0100
+++ ipcs.c 2010-12-03 09:33:08.0 +0100
@@ -168,7 +168,7 @@
assert(s->ref_count > 0);
free_it = qb_atomic_int_dec_and_test(&s->ref_count);
if (free_it) {
-
Hi all,
I try to track confdb changes,i.e. I want to get notified about cman config
changes.
corosync-objctl -t cluster
Type "q" to finish
object_deleted>cluster
I only get one "confdb_object_delete_change_notify " notification
after running "cman_tool version -r -S". Is that expected behavio
> If I use libpb, it creates a second thread pool, so I end up running twice as
> many threads?
Oh, I guess this only matters if I write a corosync service. Simply ignore
that question.
- Dietmar
___
Openais mailing list
Openais@lists.linux-foundatio
> I'd recommend using libqb if you want to use epoll with ipc - its already
> available. Its far improved and where we are headed in corosync 2.0.
Thanks for that hint. I want to use it inside an application which
use CPG (and corosync 1.2), so I thought if I use use libcoroipcs/ipcc
I share t
> No, you should either get a LEAVE or a PROCDOWN but not both. I have
> verified this is indeed the case on master. Could you describe your test case
> in more detail?
Well, I will try to extract a simple test which shows the behavior next week.
- Dietmar
_
Hi all,
is there some example code on howto use libcoroipcs with epoll? Is that
expected to work? Or do I need to use poll() instead?
- Dietmar
___
Openais mailing list
Openais@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/list
Hi all,
i just detected that I get 2 CPG leave message for the same configuration.
for example I get:
members: 1 2
left: 3 (CPG_REASON_LEAVE)
and then:
members: 1 2
left: 3 (CPG_REASON_PROCDOWN)
Is that expected behavior?
- Dietmar
___
Ope
> Our project is focused around one goal: providing a small footprint
> (10kloc) highly available block storage area for virtual machines
> optimized for Linux data-centers. Our plans don't depend on SAN
> hardware, software, hardware fencing devices, or any other hardware
> then
> is commonly ava
> The long term goal of this work is to enable a replicated structured
> memory-based key-value storage that maintains consistency after a merge
> from a network partition. This allows IPC speed reads, and network
> speed writes of key/value pairs with full availability of key/value
> data on all
Do you have an idea whats the best place to implement self fencing? Can
we simply use softdog inside the quorum service (to trigger a reboot
when we lose quorum? or is that too simple? Or is fenced the better place?
- Dietmar
> > But what I've heard so far is that many users do not understand
> >
sync 1.2.0.
Is somebody able to reproduce that issue?
- Dietmar
> Regards,
> Honza
>
> Dietmar Maurer wrote:
> > Just found the following commit:
> >
> >
> http://git.fedorahosted.org/git/cluster.git?p=cluster.git;a=commitdiff;
> h=bcc5fdef8473d99399c624a7b
d after IPC is finished.
> >
> > Maybe it is bug. Do you have any reproduces?
> >
> > Thanks,
> > Honza
> >
> > Dietmar Maurer wrote:
> > >> Inside my CPG application, The confchg callback is called with
> > 'dead'
> > &
d after IPC is finished.
>
> Maybe it is bug. Do you have any reproduces?
>
> Thanks,
> Honza
>
> Dietmar Maurer wrote:
> >> Inside my CPG application, The confchg callback is called with
> 'dead'
> >> members:
> >>
> >
> > Currently there are two choices 1) power fencing 2) no fencing.
>
> Wll, you could configure both.
> But you'd end up with the node being power fenced after committing
> sepuku, which wouldn't be ideal.
But we can do:
1.) use regular fencing if configured
2.) If there is no fencing conf
> > There are thousands of interactions with power fencing and every one
> > of them needs to work perfectly for power fencing to work.
>
> Thats not the problem.
> Its the false positives you need to worry about (devices that report
> success when power fencing failed).
>
> When power fencing fa
> Dietmar,
> process *should* be removed after IPC is finished.
>
> Maybe it is bug. Do you have any reproduces?
>
It happens quite often here, but I have no easy way to reproduce it.
Where is that 'IPC' code responsible for removing the process (maybe
I can debug myself)?
- Dietmar
_
> Inside my CPG application, The confchg callback is called with 'dead'
> members:
>
> [debug] cpg member node 3 pid 1132
> [debug] cpg member node 3 pid 14640
>
> for example process 1132 does not exists any longer on node 3. Any idea
> what
> can cause such 'ghost' entries?
If I run corosync-c
Hi all,
I currently test cman (3.0.7) and corosync 1.2.0 on a debian squeeze box.
Inside my CPG application, The confchg callback is called with 'dead' members:
[debug] cpg member node 3 pid 1132
[debug] cpg member node 3 pid 14640
for example process 1132 does not exists any longer on node 3.
Hi all, I just found a whitepaper from XenServer - seem they implement some
kind of self-fencing:
-text from XenServer High Availability Whitepaper---
The worst-case scenario for HA is the situation where a host is thought to be
off-line but is actually
still writing to the shared storage
> On 19/02/10 01:30, Steven Dake wrote:
> > On Thu, 2010-02-18 at 16:04 -0800, Alan Jones wrote:
> >> Friends,
> >> Is there any (undocumented) configuration option for corosync to
> list
> >> the nodes in the cluster similar to heartbeat?
> >> Alan
> >>
> >
> > No but clearly this is a gap. This
> >>> But that can't work. corosync (when using cman) needs cluster.conf
> to
> >>> start up. So if the filesystem depends on corosync you have a
> catch-
> >> 22
> >>> situation: the filesystem can't be mounted because corosync isn't
> >>> running
> >>
> >> That is no problem. You can mount the fi
> -Original Message-
> From: Dietmar Maurer
> Sent: Donnerstag, 28. Jänner 2010 11:31
> To: 'Christine Caulfield'
> Cc: openais@lists.linux-foundation.org
> Subject: RE: [Openais] quorum service question
>
> > But that can't work. corosy
> But that can't work. corosync (when using cman) needs cluster.conf to
> start up. So if the filesystem depends on corosync you have a catch-22
> situation: the filesystem can't be mounted because corosync isn't
> running
That is no problem. You can mount the filesystem, but it is read-only
as lo
> -Original Message-
> From: Christine Caulfield [mailto:ccaul...@redhat.com]
> Sent: Donnerstag, 28. Jänner 2010 10:49
> To: Dietmar Maurer
> Cc: openais@lists.linux-foundation.org
> Subject: Re: [Openais] quorum service question
>
> On 28/01/10 09:38, Dietmar M
> >>> But who calls 'cman_tool version' to actually sync the config?
> >>
> >> You do ;-)
> >
> > Really? I have to connect to each node and call 'cman_tool version'?
>
> No, just one one node.
But what if the cluster is partinioned?
___
Openais mailin
> >> It's called ccs_sync and is part of the ricci package.
> >>
> >> cman_tool will automatically call this (if it's installed) when you
> >> update the version number and call 'cman_tool version'.
> >
> > Ok, found that code. Seems that ricci does the following:
> >
> > - generate new config file
> > I just tried to find the code which syncs cluster.conf to all nodes
> > in current Red Hat Cluster, but was not able to find it. Please can
> someone
> > give me a hint what source files implement that? Seems there is no
> ccsd in
> > cluster3?
> >
> >
>
> It's called ccs_sync and is part of t
> > >> The quorum service really works best if you have a common
> > configuration
> > >> system so that the values are the same on all nodes.
> > >
> > > The question is how to do that. Quorum setting are read from
> > /etc/corosync/corosync.conf, but there is no way to write that file -
> > the c
> >>> these weren't available for openais whitetank, so we have a custom
> >>> version that we will "soon" replace with the one from corosync.
> >>
> >> Great! I guess you will use votequorum?
> >
> > It would be configurable I imagine.
>
>
> If all you need to know is "do we have quorum?" then y
> On Wed, 2009-11-18 at 14:01 +0100, Dietmar Maurer wrote:
> > I just noticed that cpg_join return CS_ERR_TRY_AGAIN, but it joins
> the group anyways. Is that expected?
> >
> No. How do you know it joined the group?
I got a configuration change event.
> How do you re
> > So quorum setting should be the same on all nodes. What other
> settings should be the same on all nodes? Are the totem settings local
> or should they be the same on all nodes?
> >
> > - Dietmar
> >
>
> Each node's config file should be an exact copy, including totem
> network
> settings.
I
I just noticed that cpg_join return CS_ERR_TRY_AGAIN, but it joins the group
anyways. Is that expected?
Also, I get that CS_ERR_TRY_AGAIN many times, so how should I call it:
while ((result = cpg_join(mdb->cpg_handle, &mdb->cpg_group_name)) ==
CS_ERR_TRY_AGAIN)
sleep (1);
Or is there
I am trying to answer that myself,
> > > That way you use 2 different services - votequorum an CPG. Is there
> > > really a defined order between those messages?
> >
> > There are no quorum messages, but correlating events between cpg and
> > quorum is a problem,
>
> I think about injecting a mes
> >> The quorum service really works best if you have a common
> configuration
> >> system so that the values are the same on all nodes.
> >
> > The question is how to do that. Quorum setting are read from
> /etc/corosync/corosync.conf, but there is no way to write that file -
> the confdb interfac
> >> We hope there will be more in the future to fulfil various needs.
> >
> > Which is used by pacemaker?
>
> neither, yet.
> these weren't available for openais whitetank, so we have a custom
> version that we will "soon" replace with the one from corosync.
Great! I guess you will use votequoru
> Currently there are three quorum providers:
>
> votequorum
> YKD
> testquorum
I can find votequorum and testquorum, but where is YKD (what source files)?
> testquorum is (as it says) just a test - it can flip the quorum state
> based on an objdb value, but the other two provide quorum services
> > That way you use 2 different services - votequorum an CPG. Is there
> > really a defined order between those messages?
>
> There are no quorum messages, but correlating events between cpg and
> quorum is a problem,
I think about injecting a message into the CPG when quorum changes. But this
> >> The quorum service really works best if you have a common
> configuration
> >> system so that the values are the same on all nodes.
> >
> > The question is how to do that. Quorum setting are read from
> /etc/corosync/corosync.conf, but there is no way to write that file -
> the confdb interfac
> > So what is the suggested way to block some CPG operation when quorum
> is lost (how to integrate quorum into an CPG application)?
> >
>
> Use the quorum library, and check the cluster is quorate before sending
> messages. You don't need to call the quorum_getquorate() call every
> time, you ca
> > But AFAIK this information is local (not copied to other nodes). And
> the votequorum seems to add new nodes dynamically. So how can I make
> sure that all nodes in the cluster use the same values? I guess it is
> important that all nodes start with the same value?
> >
> > Or is it better to no
I can see the difference in the API, but what is exactly the difference - both
are API into the votequorum service?
- Dietmar
___
Openais mailing list
Openais@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/openais
Hi all,
I just notice that cpg_leave hangs forever when I stop the corosync daemon. Is
that expected behavior? If so, is it safe to just call cpg_finalize - does that
free all resources?
- Dietmar
___
Openais mailing list
Openais@lists.linux-foundati
> > So CFG, CPG and VOTEQUORUM works normal when the cluster is
> inquorate? But EVS does not?
> >
>
>
> Yes, that's how it currently is - I suspect that all openais services
> are blocked too.
>
> It might be the case that some other subsystems need to operate when
> there is no quorum - when q
> > What operations are blocked - I can't find that in the source code?
> >
>
> See in main.c
>
> static int corosync_sending_allowed()
# grep CS_LIB_ALLOW_INQUORATE services/*.[ch]
services/cfg.c: .allow_inquorate=
CS_LIB_ALLOW_INQUORATE,
services/confdb.c: .allow_
After some, testing, I noticed that I can't reproduce the bug anymore, don't
know if this is good or bad ;-)
> -Original Message-
> From: Fabio Massimo Di Nitto [mailto:fabbi...@fabbione.net]
> Sent: Freitag, 30. Oktober 2009 06:10
> To: Dietmar Maurer
> Cc: Chr
And what is the rationale behind this check:
"if lowest processor identifier of the old ring in the new ring"
it is used in the checkpoint recovery algorithm:
http://www.openais.org/doku.php?id=dev:partition_recovery_checkpoint:checkpoint
- Dietmar
> > but with
> >
> > # ./testvotequorum2 5
> > votequorum_qdisk_getinfo error 12: OK
> > qdisk votes 4
> > state0
> > name QDISK
> >
> > the cluster freezes (corosync does not respond anymore)?
> >
> > (using corosync-1.1.2 on debian lenny).
> >
>
>
> I don't know what would be c
> I still recommend using CPG for all use since it receives the most
> testing.
Ah, OK - many thanks for your help.
- Dietmar
___
Openais mailing list
Openais@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/openais
> > And what is the difference between EVS and CPG?
> >
>
> EVS is a passthrough without process id membership. cpg provides
> process group membership meaning that both node id and process id
> information is given for membership information.
So if I plan to only run one instance per node its b
Can anyone else reproduce that error?
Any suggestion how to debug such thing?
> -Original Message-
> From: Christine Caulfield [mailto:ccaul...@redhat.com]
> Sent: Donnerstag, 29. Oktober 2009 15:31
> To: Dietmar Maurer
> Cc: openais@lists.linux-foundation.org
> Sub
Btw, when I start
# testvotequorum2
votequorum_qdisk_getinfo error 12: OK
votequorum_qdisk_getinfo error 12: OK
but with
# ./testvotequorum2 5
votequorum_qdisk_getinfo error 12: OK
qdisk votes 4
state0
name QDISK
the cluster freezes (corosync does not respond anymore)?
(usin
> Actually it does block some API operations when there is no quorum - or
> it should.
What operations are blocked - I can't find that in the source code?
- Dietmar
___
Openais mailing list
Openais@lists.linux-foundation.org
https://lists.linux-founda
Hi all,
I am just testing the quorum service and have some basic questions. First, I
use the following configuration in corosync.conf:
quorum {
provider: corosync_votequorum
expected_votes: 3
votes: 1
}
But AFAIK this information is local (not copied to other nodes). And
Hi all,
The evs_overview man page says:
>Virtual Synchrony allows the current configuration to be used to make
>decisions in partitions
>and merges. Since the configuration is sent in the stream of messages to the
>application, the
>application can alter its behavior based upon the configuratio
> This is how I think it could work, based on the idea that developers
> use
> trunk and integration use flatiron.
>
> - bug is reported/fixed against trunk:
> * verify right away if the bug affects Flatiron or not.
> * If it does, cherry pick the fix in Flatiron for integration people
> t
> -Original Message-
> From: openais-boun...@lists.linux-foundation.org [mailto:openais-
> boun...@lists.linux-foundation.org] On Behalf Of Dietmar Maurer
> Sent: Dienstag, 18. August 2009 07:54
> To: sd...@redhat.com; open...@lists.osdl.org
> Subject: Re: [Opena
We need to install corosync without user interaction, so that solution is even
worse than the previous behavior. Can't we use /dev/urandom instead (AFAIK even
ssh uses that to generate private keys)?
- Dietmar
> -Original Message-
> From: openais-boun...@lists.linux-foundation.org [mail
> When I start openais+pacemaker on my virtual machines (vbox) it uses a
> lot of
> CPU power. top shows 30 - 50% of CPU usage only for ais.
>
> The virtual manchines are not the slowest (~4000 bogomips). Please
> could
> anybody explain me the problem? When I use heartbeat as the cluster
> stack
> As I mentioned in another mail, some apps may allow re-syncing without
> a complete node reset or rejoining the cpg. Other apps (like dlm and
gfs
> that I work on) require a full node reset to re-sync, because they
export
> their state through the dlm and gfs and up into random
But exporting s
> To replicate state using virtual synchrony / cpg you must have agreed
> ordering
> of messages and configuration changes, which you don't get by using
> separate
> cpgs for each.
It uses only one cpg to send (application) messages, so you still have
the VS guarantee. The other cpg is only used t
> The key point is that two cpg members (A and B) with different event
> histories
> will have inconsistent state, assuming that the state is derived from
> the
> history. So, after they are partitioned and merged, A needs to
replace
> its
> own state/history with B's or v.v.
Right. That is the i
> > > And even worse, section contains data from different partitions
> (old
> > > data mixed with new one)? And there is no notification that such
> things
> > > happens?
>
> That ckpt behavior is nonsensical for most real applications I'd
wager.
> I'm going to have to go check whether my apps ar
> Merging/reconciling a partition is *fundamentally* not possible for
any
> application that requires virtual synchrony. You cannot rewrite
> history,
But I can re-sync the state.
> and VS is all about event history. It has nothing to do with
mechanism or
> design, it has everything to do with
> If your application requires virtual synchrony, then the only way to
> handle
> partitions is to terminate one part (and force it to restart with no
> history)
> to preserve a common history among the remaining nodes. Merging
> doesn't make
> sense because the VS history has already diverged.
A
> Here is an example CPG application which shows how to replicate state
> between multiple processes on multiple machines. It is just a start,
> but have a look. If you have interest in CPG, it may help you
> understand the basics of the API.
Interesting, especially the use of 2 process groups.
> It took us a long time to develop some patterns around application
> programming over VS. But there is a good book out there that can help:
> "Building Secure and Reliable Network Applications" by Ken Birman
> (whose Cornell research team invented VS)
Thanks, will take a look at that book.
- Di
> The SA Forum doesn't consider at all how to handle partitions in a
> network or at least not very suitably (up to designer of SA Forum
> services). They assume that applications will be using the AMF, and
> rely on the AMF functionality to reboot partitioned nodes (fencing) so
> this condition d
> > Is there any guarantee that all sections inside a recovered
> checkpoint
> > are from the same cluster partition (I can't see such restriction in
> the
> > algorithm)?
> >
>
> no
>
> The algorithm will merge sections created in both partitions into the
> single checkpoint.
At least the SA Fo
> An older version of the algorithm is described here:
>
http://www.openais.org/doku.php?id=dev:partition_recovery_checkpoint:ch
> eckpoint
>
> It has been updated to deal with some race conditions, but the
document
> is pretty close.
>
> As you can see, designing the recovery state machine is co
> An older version of the algorithm is described here:
>
http://www.openais.org/doku.php?id=dev:partition_recovery_checkpoint:ch
> eckpoint
>
> It has been updated to deal with some race conditions, but the
document
> is pretty close.
>
> As you can see, designing the recovery state machine is co
> In my opinion to merge states reliably, the designer of the
application
> has to do the hard work of designing a synchronization protocol that
> happens each time a configuration change occurs.
Yes. (That is why I am confused about the checkpoint service - how can
that service
merge states relia
> Yes, forcing the losers to reset and start from scratch is a must, but
> we end up doing that a layer above corosync. That means the losers
often
> reappear again through corosync/cpg prior to being forced out.
Are you talking about an implementation bug, or a 'bubbling idiot' which
simply joi
> > > > like a 'merge' function? Seems the algorithm for checkpoint
> recovery
> > > > always uses the state from the node with the lowest processor
id?
> > > >
> > > Yes that is right.
> >
> > So if I have the following cluster:
> >
> > Part1: node2 node3 node4
> > Part2: node1
> >
> > Let assume
> Once a partition exists, a merge back together doesn't change the fact
> that
> the disagreement has already occured (at partition time) and that
> disagreement
> can only be resolved (to maintain VS) by killing nodes that don't
agree
> with
> one version of the history.
Sure, but the whole prob
> > like a 'merge' function? Seems the algorithm for checkpoint recovery
> > always uses the state from the node with the lowest processor id?
> >
> Yes that is right.
So if I have the following cluster:
Part1: node2 node3 node4
Part2: node1
Let assume Part1 is running for some time and has gath
Lest assume the cluster is partitioned:
Part1: node1 node2 node3
Part2: node4 node5
After recovery, what join/leave messaged do I receive with a CPG:
A.) JOIN: node4 node5
or
B.) JOIN: node1 node2 node3
or anything else?
- Dietmar
___
Openais mail
> Check out:
> http://www.openais.org/doku.php?id=dev:paritition_recovery
>
> Instead of using the token callback method, you could write your own
> methodology for executing the state machine.
Ah, OK - I think that is what I already do. What is miss is something
like a 'merge' function? Seems th
> You might try taking a look at exec/sync.c
>
> it is a synchronization engine. Basically it takes configuration
> changes into account to call sync_init, sync_process, sync_abort, or
> sync_activate. These 4 states then activate the new dato a model.
> This
> could easily be done as an addon a
> Cool idea if it wasn't totally integrated with CPG but instead some
> external service which people could use in addition to CPG.
Thats the plan.
My current problem is the state merge function. There should be a
standard way to merge state (when the user does not want to provide his
own merge
> You might try taking a look at exec/sync.c
>
> it is a synchronization engine. Basically it takes configuration
> changes into account to call sync_init, sync_process, sync_abort, or
> sync_activate. These 4 states then activate the new dato a model.
> This
> could easily be done as an addon a
> > > Yes. I'm not sure if a generic service for this would be used
much
> or
> > > not...
> > > maybe.
> >
> > Btw, how large is an average checkpoint in dlm? Just wonder how much
> > data needs to be transferred.
>
> I've never measured, but it's a trivially small amount of data.
Around
> 32
>
> > When a new node joins, CPG immediately change mode to
> > DFSM_MODE_SYNC. Then all members send their state.
> >
> > When a node received the states of all members, it computes the
new
> > state by merging all received states (dfsm_state_merge_fn), and
> > finally switches mode to DFS
states (dfsm_state_merge_fn), and
finally switches mode to DFSM_MODE_WORK.
Does that make sense?
- Dietmar
> On Thu, Apr 09, 2009 at 09:00:08PM +0200, Dietmar Maurer wrote:
> > > If new, normal read/write messages to the replicated state
continue
> while
> > > the new
If I send 2 messages with CPG_TYPE_AGREED:
node1: cpg_mcast_joined (A)
node1: cpg_mcast_joined (B)
Is there any guarantee that they arrive in the same order (A before B)?
- Dietmar
___
Openais mailing list
Openais@lists.linux-foundation.org
https://li
> guarantees you seek, and if it doesn't, it is defective. The only
> exception might be if the new process reuses the same PID since the
> pid/nodeid/group are the uniqifiers and if pid is the same, there is
no
> way to detect the new process (and remove the old one).
PID reuse happens more ofte
> > Ah, that probably works. But can lead to very high memory usage if
> traffic
> > is high.
>
> If that's a problem you could block normal activity during the sync
> period.
wow. that 'virtual synchrony' sound nice first, but gets incredible
complex soon ;-)
>
> > Is somebody really using tha
> 1. Have an old cpg member (e.g. the one with the lowest nodeid) send
> messages
> containing the state to the new node after it's joined. These "sync
> messages"
> are separate from the messages used to read/write the replicated state
> during
> normal operation.
This is not bullet proof. State
> > > need for locks. An example of why not is creation of a resource
> > called
> > > "datasetA".
> > >
> > > 3 nodes:
> > > node A sends "create datasetA"
> > > node B sends "create datasetA"
> > > node C sends "create datasetA"
> > >
> > > Only one of those nodes create dataset will arrive firs
> What I recommend here is to place your local node id in the message
> contents (retrieved via cpg_local_get) and then compare that nodeid to
> incoming messages
Why do you include the local node id into the message? I can compare the
local node id with the sending node id without that, for examp
1 - 100 of 113 matches
Mail list logo
