Another node: in general, would it make sense to use shorter names ? E.g. instead of ** New view: [jdg-perf-01-60164|9] [jdg-perf-01-60164, | jdg-perf-01-24167, jdg-perf-01-53841, jdg-perf-01-39558, | jdg-perf-01-8977, jdg-perf-01-49115, jdg-perf-01-24774, | jdg-perf-01-5758, jdg-perf-01-37137, jdg-perf-01-45330, | jdg-perf-01-24793, jdg-perf-01-35602, jdg-perf-02-7751, | jdg-perf-02-37056, jdg-perf-02-50381, jdg-perf-02-53449, | jdg-perf-02-64954, jdg-perf-02-34066, jdg-perf-02-61515, | jdg-perf-02-65045 ...] we could have ** New view: [1|9] [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, ...] This makes reading logs *much* easier than having those long names.

If we wanted the host name to be part of a cluster name, we could use the alphabet, e.g. A=jdk-perf-01, B=jdg-perf-02: ** New view: [A1|9][A1, A2, A3, B4, B6, C2, C3, ...] This is of course tied to a given host naming scheme. But oftentimes, host names include numbers, so perhaps we could use a regexp to extract that number and use it as a prefix to the name, e.g. cluster-01 first instance: 1-1 cluster-02 2nd instance: 1-2 etc. Thoughts ?

On 3/4/13 8:43 AM, Radim Vansa wrote: > Just a small sidenote: if you want to print full view (not just first 20 nodes and ellipsis after that), use -Dmax.list.print_size=cluster_size > > Radim > > ----- Original Message ----- > | From: "Shane Johnson" <shjohnso(a)redhat.com&gt; > | To: infinispan-internal(a)redhat.com > | Sent: Friday, March 1, 2013 5:52:17 PM > | Subject: Re: [infinispan-internal] Unstable Cluster > | > | The JGroups cluster appeared stable, at first. However, I did notice > | that the logs looked a little bit different on one machine / > | instance. I'm not sure if that means anything or not. > | > | Machine 1 / Instance 1-12 > | > | ** New view: [jdg-perf-01-60164|9] [jdg-perf-01-60164, > | jdg-perf-01-24167, jdg-perf-01-53841, jdg-perf-01-39558, > | jdg-perf-01-8977, jdg-perf-01-49115, jdg-perf-01-24774, > | jdg-perf-01-5758, jdg-perf-01-37137, jdg-perf-01-45330, > | jdg-perf-01-24793, jdg-perf-01-35602, jdg-perf-02-7751, > | jdg-perf-02-37056, jdg-perf-02-50381, jdg-perf-02-53449, > | jdg-perf-02-64954, jdg-perf-02-34066, jdg-perf-02-61515, > | jdg-perf-02-65045 ...] -- Bela Ban, JGroups lead (http://www.jgroups.org) _______________________________________________ infinispan-dev mailing list infinispan-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/infinispan-dev

On Mon, Mar 4, 2013 at 10:28 AM, Bela Ban <bban(a)redhat.com <mailto:bban@redhat.com>> wrote: Another node: in general, would it make sense to use shorter names ? E.g. instead of ** New view: [jdg-perf-01-60164|9] [jdg-perf-01-60164, | jdg-perf-01-24167, jdg-perf-01-53841, jdg-perf-01-39558, | jdg-perf-01-8977, jdg-perf-01-49115, jdg-perf-01-24774, | jdg-perf-01-5758, jdg-perf-01-37137, jdg-perf-01-45330, | jdg-perf-01-24793, jdg-perf-01-35602, jdg-perf-02-7751, | jdg-perf-02-37056, jdg-perf-02-50381, jdg-perf-02-53449, | jdg-perf-02-64954, jdg-perf-02-34066, jdg-perf-02-61515, | jdg-perf-02-65045 ...] we could have ** New view: [1|9] [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, ...] This makes reading logs *much* easier than having those long names. Yes and no... I sometimes find it useful to have a somehow longer name, as searching/filtering for a node in the log is pretty much impossible with a name like "1".

I also think we need the random number at the end so that we can debug problems with node restarts. In JDG/AS7 they don't add a random number, and it was very difficult to see what was happening when a node name appeared twice in the consistent hash. But we could make the random number shorter.

If we wanted the host name to be part of a cluster name, we could use the alphabet, e.g. A=jdk-perf-01, B=jdg-perf-02: ** New view: [A1|9][A1, A2, A3, B4, B6, C2, C3, ...] This is of course tied to a given host naming scheme. But oftentimes, host names include numbers, so perhaps we could use a regexp to

that number and use it as a prefix to the name, e.g. cluster-01 first instance: 1-1 cluster-02 2nd instance: 1-2 etc. Thoughts ? Are you thinking of an automatic way of assigning a letter+digit combination to a node on startup? We also use the node name for some other stuff (e.g. thread names), so I'm not sure if it's feasible to wait until we have connected to the JGroups cluster to set the node name dynamically. For RadarGun we could use a static system where we configure a node name for each slave and then RadarGun passes the node name to Infinispan via a system property. No Infinispan changes required (except perhaps making the random number in the node name optional.)

On 3/5/13 3:30 PM, Erik Salter wrote: > Hi guys, > > Keep in mind that some of your customers may have built queries and indexes > on cluster names on top of very expensive analytics engines. Well, if they use their own naming (JChannel.setName()), no problem.

But I've said (many times) that relying on node names is a *bad thing* ! Node names are syntactic sugar, and there may not be a name associated with a node, and then it has to be fetched dynamically, using an ARP like protocol.

If someone wanted to add information to an address, then the way to do it would be to use an AddressGenerator and return subclasses of UUID, e.g. PayloadUUID.

> If this discussion is limited in scope to internal applications, that's > fine. If not, having done debugging of issues on live customer sites, I > think it's fine the way it is. >

> Erik > > -----Original Message----- > From: infinispan-dev-bounces(a)lists.jboss.org > [mailto:infinispan-dev-bounces@lists.jboss.org] On Behalf Of Bela Ban > Sent: Tuesday, March 05, 2013 1:50 AM > To: infinispan-dev(a)lists.jboss.org > Subject: Re: [infinispan-dev] [infinispan-internal] Unstable Cluster > > > > On 3/4/13 6:35 PM, Dan Berindei wrote: >> >> On Mon, Mar 4, 2013 at 10:28 AM, Bela Ban <bban(a)redhat.com >> <mailto:bban@redhat.com>> wrote: >> >> Another node: in general, would it make sense to use shorter names ? >> E.g. instead of >> >> ** New view: [jdg-perf-01-60164|9] [jdg-perf-01-60164, >> | jdg-perf-01-24167, jdg-perf-01-53841, jdg-perf-01-39558, >> | jdg-perf-01-8977, jdg-perf-01-49115, jdg-perf-01-24774, >> | jdg-perf-01-5758, jdg-perf-01-37137, jdg-perf-01-45330, >> | jdg-perf-01-24793, jdg-perf-01-35602, jdg-perf-02-7751, >> | jdg-perf-02-37056, jdg-perf-02-50381, jdg-perf-02-53449, >> | jdg-perf-02-64954, jdg-perf-02-34066, jdg-perf-02-61515, >> | jdg-perf-02-65045 ...] >> >> >> we could have >> ** New view: [1|9] [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, >> 16, 17, 18, 19, 20, ...] >> >> This makes reading logs *much* easier than having those long names. >> >> >> Yes and no... I sometimes find it useful to have a somehow longer >> name, as searching/filtering for a node in the log is pretty much >> impossible with a name like "1". > > > Yes, however if you have context it's not an issue, e.g. in JGroups I > oftentimes prefix my messages with name:. So, for example, if you're looking > for unicast traffic received by 5, this would work "5: <--" as grep > argument. > > I agree this isn't useful when you want to follow the address of a member > throughout the log, and all protocols. > >> I also think we need the random number at the end so that we can debug >> problems with node restarts. In JDG/AS7 they don't add a random >> number, and it was very difficult to see what was happening when a >> node name appeared twice in the consistent hash. But we could make the >> random number shorter. > > > Would maintaining a base name (A) and then incrementing a short help ? > E.g. A1, when restarted A2 ? The problem is that we'd have to store the > number on disk... > > >> If we wanted the host name to be part of a cluster name, we could use >> the alphabet, e.g. A=jdk-perf-01, B=jdg-perf-02: >> >> ** New view: [A1|9][A1, A2, A3, B4, B6, C2, C3, ...] >> >> This is of course tied to a given host naming scheme. But oftentimes, >> host names include numbers, so perhaps we could use a regexp to > extract >> that number and use it as a prefix to the name, e.g. >> cluster-01 first instance: 1-1 >> cluster-02 2nd instance: 1-2 >> etc. >> >> Thoughts ? >> >> >> Are you thinking of an automatic way of assigning a letter+digit >> combination to a node on startup? We also use the node name for some >> other stuff (e.g. thread names), so I'm not sure if it's feasible to >> wait until we have connected to the JGroups cluster to set the node name >> dynamically. >> >> For RadarGun we could use a static system where we configure a node name >> for each slave and then RadarGun passes the node name to Infinispan via >> a system property. No Infinispan changes required (except perhaps making >> the random number in the node name optional.) > > > I think this would already help; for apps / tests that we control, we > should strive to make names as short as possible. I often find myself > awking and seding a log file before I even look at it, as names are way > too long. > > Would it help if a base name could be set in a channel, e.g. > channel.setBaseName("A") and then we maintain a file A in temp storage > to which we write 1 (first time, file doesn't exist) ? Next time we'd > read the file (which matches the base name) and write the incremented > number (2) ? So we'd have member A1, then A2 and so on... > > Thoughts ? > -- Bela Ban, JGroups lead (http://www.jgroups.org) _______________________________________________ infinispan-dev mailing list infinispan-dev(a)lists.jboss.org https://lists.jboss.org/mailman/listinfo/infinispan-dev