MRU multipath failover does not work, hangs host. Fixed does not work on CX500?

Similar Questions

• The NIC failover does not work in LBT mode

  Dear community,
  

  
  

  We met a strange behavior in our network environment while testing our new Cisco 4500 switches x 10g. If you have a TCP Session (example JJ | netcat) data transfer between two virtual machines on different host esx failover does not work on the VM receiver if a physical connection fails.

  The test configuration is as follows:

  |    | - 4 - switch 1 vmnic - vmnic 4-|    |

  VM1 | ESX1 |                  | VSS |                    | ESX2 |-VM2

  |    | - 5 - switch 2 vmnic - vmnic 5-|    |

  More details:

  • The switches are configured as a cluster VSS. New version of the software
  • ESX1/2 are running the latest version 5.5
  • The two vmnic is active on a DV-switch in LBT mode
  • Exchanges are configured with "Port based on the physical NIC load."
  • VM1 and VM2 are invited linux, NICS are vmxnet3

  Steps to reproduce:

  1. Beginning listening mode on VM2 (netcat-l Pei 12345 |) JJ = / dev/null)
  2. Start the transfer on VM1 (dd if = / dev/zero bs = 1 M | netcat VM2 12345)
  3. Check which link is used (in my example vmnic5 on ESX2)
  4. Unplug the ESX2 vmnic5

  What you observe:

  • DD-traffic dies immediately
  • If you kill the dd on machines (it does not end by itself) and start all over again, it works immediately
  • If you ping VM1 VM2 or another system, it works
  • If you plugged vmnic5 again once the virtual machine is switched to new vmnic5-> if you didn't kill dd will continue

  Now to test the reverse:

  1. Beginning listening mode on VM2 (netcat-l Pei 12345 |) JJ = / dev/null)
  2. Start the transfer on VM1 (dd if = / dev/zero bs = 1 M | netcat VM2 12345)
  3. Check which link is used (in my example vmnic4 on ESX1)
  4. Disconnect vmnic1 ESX1

  What you observe:

  • Failover works like a charm
  • Failback works like a charm

  Conclusion:

  • Only the virtual machine at the front desk has a problem
  • Only an already established TCP session behaves badly, work of new sessions very well (as does ping, no losses)
  • If you use LACP place LBT on uplink the problem disappeared
  • Same problem if you connect two links to Switch1
  • Same problem if you use a physical box with linux and LACP as system
  • I did the same test with two c3750x (no cluster VSS) and has not encountered the problem

  Any ideas? It's really a good one.

  Thank you!

  Sascha

  
  

  
  

  Update:

  To make it clear, it doesn't have the transmission of UDP.

  Post edited by: sewert1

  I did some further testing and also involved cisco. There is no problem with the switches and network in general.

  Turns out he works with Windows clients, but it does not work with Linux. But it works with Linux with adapters e1000 and e1000e but not with vmxnet3.

  So I did some tests and learned that the vmxnet3 kernel module must be at least version 1.3. Version 1.1 (SLES 11 default) or 1.2 (default SLES12) does not work.

  (./vmware-install.pl--clobber-noyau-modules = vmxnet3)

• "The SP failover is not supported. Single SP port failover is supported", that means

  In the compatibility list VMware ESXi 4.x or promise VessRAID SAN iSCSI has a note: 'the SP failover is not supported. SP port failover is supported only"means MS obviously storage path', however I am not sure what is meant by this note, since everything I can find when searching for documents of VMware for"port failover"is a reference to HBA that hide the failover of the VMware kernel, which is not a property of the target iSCSI SAN or VMware aside initiator parameters.

  That means this note, what settings VMware cannot be used and what are those who can.  Promise of support is often useless, so asking them is not likely to get a direct answer.

  Maybe the promise VessRAID iSCSI SAN has only a processor unique storage (SP) with several ports storage/controller?

  When you look at their website, it seems that the c-series is single controller only... So obviously it is not able to switch to another storage processor if the primary fails...

  / Rubeck

• failover - does not change clerks

  Hello

  I ve read some docs on cookies, sessions, passivation, etc. and came across the following piece:

  To check, try the following: create the simple application of ADF business components and the web client to it. Define the HttpSession timeout to 1 minute and make sure that the switch is turned on, run your client application, create a new line, or change a value but do not commit (or rollback). Now wait for more then a minute to let the HttpSession timeout. Note that if in your ViewController project you add a property to the Java virtual machine on the Panel "Runner" like -Djbo.debugoutput=console you will actually see when your HttpSession time in the debug trace information. Now, try to continue working. You will see that your State has been preserved as if the HttpSession never expired. You will be able to see all your changes that have not been made yet and you will still be able to commit as well. Now try the same scenario, but now, with tipping off. You will see that after the time-out period, your wait state is lost. You will then have to start from the point where you last committed your changes.

  to understand the State of ADF Business components management feature.

  What I understand here is possible when a user session expires and changes that have not committed will be always 'available' for the user to commit when you connect again, I am rigth?

  If Yes, I followed the instructions and it didn t happened as indicated.

  If not, is it possible to do this? Accordingly to each validated user to present its not commit changes?

  Jdev 12 c

  Kind regards

  Carlos

  The doc you speak is very old. This does not mean that it bad, but the chance is there changes that cancels the content.

  Another thing to understand is that disconnection does not record the users session but frees the resources. Failover works (if it works at all the check page in ADF 12.1.3 jsf does not correctly after the session), if a session expires and the module of the application has not been recycled and the user remains on the same browser.

  What you try to do means you want to save the log of current transactions again (which can be done from another browser or pc) and work with the trade recorded again. I don't think it will work.

  Timo

• Tuxedo failover does not

  Hello

  We work with Tuxedo 11 GR 1 material for us to access an application. The configuration file uses two IP addresses from database, related to the same instance that is configured in active/passive cluster configuration. Once listed in the ubbconfig Tuxedo servers are started, the application works correctly. We made a failover test, leaving the first crux of database failure, has been denied access to the application.

  Everything worked again once the servers did an automatic restart, but after 15-20 minutes. Is it possible to reduce (if it exists) the time-out to allow automatic restart occurs when any situation crashing?

  Thank you very much

  Giuseppe.

  Hi, Giuseppe.

  The answer to your question depends on how are managed connections to the database.  If the servers are part of a transactional group, i.e. There are TMS servers associated with the Group and servers have been built with the-r switch to buildserver, Tuxedo will manage connections.  Which means that Tuxedo uses the xa_open() call to establish a connection to the database for the application, and the application should not artist all SQL CONNECT statements.  If Tuxedo receives a XA error during transaction processing, Tuxedo will automatically attempt to reestablish the connection to the database.  Also note that if you use RAC, you must set the TUXRACGROUPS environment variable and configure the database to use the DTP Services.  The DTP Service must be configured for the failover to the other instance in the database configuration file.

  On the other hand, if the server is not part of a transactional group, then all the connection of database management is left to demand and the Tuxedo is completely unaware that a database is still used.  If your code performs a SQL CONNECT and if it receives an error at some point, he probably has to reconnect with an another SQL CONNECT.  To help you in this situation, you should be able to use TAF (application transparent failover) to let the database client code tent her reconnects.

  As far as the wait time goes, I guess that your servers are suspended from a DML statement by train and finally Tuxedo kills servers due to SVCTIMEOUT.  Why they are hanging is still a stranger that I don't know how your application accesses the database or how your servers are coded.

  Kind regards

  Todd little

  Chief Architect of Oracle Tuxedo

• oci_close does not release the connection when you use DRCP

  Hi all
  
  We are currently testing the deployment of DRCP with 11g. I have the setup of this (to my knowledge), but I am facing a problem. Oci_close call does not seem to release the connection to the pool, because I expect and that is why we see similar behavior as we receive without using the DRCP.
  
  Our facility is using two instances of CCR 11.1.0.6.0 running, I use PHP 5.1.6 with installed from PECL oci8 1.3.4. DRCP pool is configured and launched, each with 100 servers max.
  
  When the Web server is idle, well, it looks like slow motion.
  
  SQL > SELECT INST_ID select, NUM_BUSY_SERVERS OF SGS$ CPOOL_STATS;
  
  INST_ID SELECT NUM_BUSY_SERVERS
  ---------- ----------------
  1 0
  2 0
  
  
  The script is as simple as that:
  
  <? PHP
  $c is oci_pconnect ('scott', 'tiger', 'IWPPOOLED');.
  $s = oci_parse ($c, ' select * from emp');
  $r = oci_execute ($s, OCI_DEFAULT);
  oci_close ($c);
  Sleep (30);
  ? >
  
  What I expected, it's that the script would be to connect to the pool, doing work for a little while and then release the connection for use by other scripts.
  
  But after I point the browser to the script, I get a 30 seconds time of loading (as expected) but the server is busy all the time, like this:
  
  SQL > SELECT INST_ID select, NUM_BUSY_SERVERS OF SGS$ CPOOL_STATS;
  
  INST_ID SELECT NUM_BUSY_SERVERS
  ---------- ----------------
  1 0
  2 1
  
  After 30 seconds of sleep, he is released and busy servers are at 0.
  
  If I load the server with ab using 256 connections:
  
  AB - n 1000000 - c 256 k - http://mywebserver/ocitest.php
  
  the pool is at the maximum and links are at a standstill:
  
  SQL > SELECT INST_ID select, NUM_BUSY_SERVERS OF SGS$ CPOOL_STATS;
  
  INST_ID SELECT NUM_BUSY_SERVERS
  ---------- ----------------
  1 95
  2 95
  
  
  My config this service network is the following:
  IWPPOOLED =
  (DESCRIPTION =
  (LOAD_BALANCE = ON)
  (FAILOVER = ON)
  (ADDRESS = (PROTOCOL = tcp) (HOST = 10.1.16.33)(PORT = 1521))
  (ADDRESS = (PROTOCOL = tcp) (HOST = 10.1.16.34)(PORT = 1521))
  (CONNECT_DATA =
  (SERVICE_NAME = IWP)
  (SERVER = POOLED)
  (FAILOVER_MODE =
  (TYPE = SELECT)
  (METHOD = BASIC)
  (RETRIES = 5)
  (TIME = 3)
  )
  )
  )
  
  The phpinfo() function will look like this:
  Active oci8 support
  Version 1.3.4
  Revision $Revision: $1.269.2.16.2.38.2.20
  Active persistent connections 1
  Active connections 1
  Oracle Instant Client Version 11.1
  Active temporary Lob support
  Support for collections enabled
  
  Directive value local value Master
  OCI8.connection_class IWPAPP IWPAPP
  OCI8.default_prefetch 100 100
  OCI8. Events on the
  OCI8.max_persistent - 1-1
  Off oci8.old_oci_close_semantics
  OCI8.persistent_timeout - 1-1
  OCI8.ping_interval - 1-1
  Off oci8.privileged_connect
  20 20 oci8.statement_cache_size
  
  
  I use the instant client for 11g
  
  Any ideas?
  
  Thank you!
  Michal

  Don't forget to use oci_free_statement ($s); See "Closing Oracle Connections" in the Underground PHP and Oracle manual. (I was just simplify this example today for the next version of the manual).

  You can also see the "dedicated optimization", where a server pooled in a non-maxed-out pool is retained (except if needed by another process PHP) assuming that the initial process of PHP can become active again. See http://www.oracle.com/technology/tech/php/pdf/php-scalability-ha-twp.pdf

  Feel free to email me (see my profile) offline if questions/data that you don't want to display.

  CJ

  Published by: cj2 on October 16, 2008 08:12

• EMC Clariion CX3 not supported in 5.1, but will it work?

  Hi all.  We are about to move to ESXI 4.1U2 5.1 in the coming weeks.  We have an old EMC Clariion CX3 - 20 c I used as a test for some time zone.  It is not our main Bay (those who are Equallogics) and is used only for testing and other uses completely non-critical.

  The HCL for 5.1 shows that nothing in the CX3 family is supported.  My questions are, it still works and if yes, are there the required settings?  I agree with it not being supported because it serves such a non-critical role, as long as I can get it connected.  Just trying to squeeze a few more life to this old Bay.

  We are connecting via the server software, without fibre channel stuff.

  Thank you!

  We have one and use it with ESXi 5.1..., it is stable with MRU and failover mode 1, no Robin available on CX3 series.

  not perform better, but as you we aren't using it for production stuff.

• HA 0 current failover host capacity

  Hi guys,.

  in this group, I am hosting big veru VM with booking at the virtual machine level.

  I have 4 x host, with 256 GB RAM each.

  HA cluster parameters "cluster tolerate 1 default host" and I have a few location with the following:

  4 Slots of the cluster

  4 used slot

  0 available accommodation

  0 slots of failover

  size of the host:

  256 mhz

  10vCPU

  132119MB

  I'm a little confused about these "zeroes" especially the failover slot...

  Does it say cluster can still tolerate 1 fault of the host or not?

  There is a time I had a good explanation of the communities, see below...

  But my question is is also used - available slot = > always the sum of the value indicated in the slot of failover?

  Example: -.

  Number of slots total = 22

  Used machines slot = 4 (depending on the number of running virtual machine)

  Available places = 7

  Failover = 11 Slots (depends on how many hosts failure cluster can tolerate) if it is 1 and we have 2 node cluster, then this is 50% of the total number of connectors. Means, that you can use according to this example, you can feed only on VM 11 are now no more than that.

  Thanks in advance.

  Daniele

  Daniele Fiore wrote:

  Eric H

  I understand, but can you answer my question below? in fact, it was an asnwer to you on a previous post

  Can you explain these values:

  Total slots in the cluster: 8

  Used slots: 4

  Places available: 2

  Failover slots 2

  Admission control policy: 1 host goes down

  Now the things are which means:

  That HA will begin only 2 virtual machines altogether? or 2 + 2 (seats + failover slots)

  And associates to 'f switch slots are zeros then cluster cannot tolerate any failure of the host' means-online HA will be restart No. VMs at all or simply HA will restart only some of them?

  Many thanks in advance,

  Daniele

  You have 8 slots to the total of your cluster

  4 of these slots are used by virtual machines

  You can always plug it into 2 more virtual machines

  And 2 slots are reserved in total as "ability" to allow a failover.

  In this case lets assume you have 4 hosts. Each guest can accommodate 2 slots. When you set the policy of admission to 1 control, he takes 2 slots slots total (amount of s to a single host) in the cluster. (8-2)

• ESX 4; evaluation license; HA does not

  Hello community VMware,

  I'm testing the new ESX 4 and demeanour HA (2 node cluster). I have trouble with vmware HA - VM not to switch (failover) to another ESX server in a cluster.

  My environment:

  hostA, hostB: 2 x ProLiant DL580 G5, each with Xeon of Intel (r) CPU X 7460 @ 2 .66GHz with 32 GB of ram

  1 x storage - HP shared EVA - two servers see this shared storage

  customer: 1 x (installed on vSphere) (client) standalone windows machine

  ESX 4 - evaluation version (on the way business licenses)

  3 servers are on the same NWs I use vSpehre on the client in order to manage the two ESX servers.  I clustered them (HA has been configured, the default values have been assigned). I have two test VMs (windows 2003) within the cluster. I tested VMotion - perfectly worked - I was able to move virtual machines from hostA to host b, and vice versa.

  Problem is that when I reboot hostA or host b (in order to test the failover) - no failover occurs. I see hostA in State "not responding", VM is in the disconnected state.  First time I tried this (after cluster configuration), I had no errors on the console. Second time I tried this I got an error of not enough resources to satisfy HA failover. I checked the HA configuration:

  The current failover capacity: 1

  Configured with failover capacity: 1

  Machine VM has only 1024 MB configured (not set limits) and it is located on the shared storage (lun 20GB, 5 GB is allocated for the system).  I've tried several things (reconfigure cluster again, clear admission control, registered between ESX/vShpere DNS, checked if I have a VMware HA between the characteristics of the license,...) - nothing helped.

  I am King of stuck here and you would be grateful and tricks that I missed (probably).

  Thank you!

  I am genius, and we have identified a problem. I would like to

  confirm that the problem is that you are

  experience.

  If the management of the network (Service Console on ESX, Case Management

  checked for port on ESXi group) is located on a public network, and there is no

  another network console or management service on a private network, it

  can lead to a false conclusion that the failed node is always online.

  prevent a tip-over.

  Please let me know if this matches your configuration. If so, a

  workaround would be to set a private network for management.

  A KB article addressiing this issue is in preparation.

  Feel free to write me with any questions.

  Marc

• Bruge ESXi ESX cluster failover host som

  VI er PT. igang med at opsaette our VMware losning close.

  VI har investeret i 2 ESX sphere servere tilsluttet et HP EVA SAN. Det var unfortunately alt hvad okonomien riahi til his jeg forsoger at strikke in losning fees med will failover manual til en ESXi host med local storage...

  Oprindeligt var tarpaulin at benytte connection vRanger VizionCore til at the Tagus backups af ESX clusteret men da der ikke er vStorage I have ESXi (free) kommer det ikke til at fungere

  His... er der nogen der har and godt til how men value I det extreme nedbrud (fx SAN' and) hurtigst possible kan mest kritiske maskiner startet op pa ESXi fa' in?

  Hejsa,

  Yes, the kan godt afvikle din vRanger connection pa ESXi hösten backe op lokalt og. Var naked and til and billigt/nature storage til one NFS backup shared.

  Best regards

  Frank Brix Pedersen

• CM failover: planned behavior

  Hello

  I did some testing of failover on my PS6000 until I put it into production.  I am trying to decide if one of my tests is behaving the way it is supposed to, and my expectations are wrong, or if 1 CM at the other failover does not work correctly.

  My configuration:

  PS6000

  2 CMs with a wiring for all ETH 4.

  Eth0-2 are on both iSCSI network.

  ETH 3 is both management network.

  Tests:

  1. If I pull the network mgt on the active CM is not failover to the other CM.  I waited for ~ 5 minutes and nothing.  This should toggle the map on whether mgt network does not work?  I'm guessing that he cannot simply use the other Eth3 on the other CM since that works in assets: liabilities.

  2. If I pull all the Eths 4, it is not (less than 5 minutes at least) will switch to another CM.

  3. If I pull the CM, then it will set the other CM/ETH.

  All this problem seems correct?  I wait long enough so that the other CM pick up?  I don't know if I have a partial loss of the ETH CM (just the map content) or not, so maybe my test here is not of weight.  Pulling the CM (a true failure and lack of card) seemed to put the other CM upwards about 1 minute.

  Thank you.

  He was named 'changeover of controllers' because it's only a failover when a controller defective and not, if you pull the cables or the stop switch set!

  To be safe against your case 2. You must crosscable a controller to two or several switches and not one.

  Against problem 1 there is only one solution, which means to connect the cable series also. Don't forget that EQL is a controller configuration "active / standby" and not an active/active.

  The most recent PSx1xx have a "horizontal port failover" but do not know if this feature works also on the Port of MGMT iam.

  Kind regards

  Joerg

• Cisco ASA 8.4 Active Failover / standby with anyconnect local CA

  Hi Friend´s

  I hope you do well! I ve got a question, hope you can help me. I ve got an ASA 5550 with version 8.4 (6), it s focusing anyconnect VPN remote access who authenticate through certificate locally generated in ASA. We´ve got an another 5550 with the same hardware and same version, and we focus on the configuration of the failover. I ve heard of network other than it s engineers may not failover configuration when the ASA doing this local. Then I ve read full failover for version 8.4 operating guide (6) and I didn t find any restrictions on the local failover and CA working together. I m tests over the next weekend, but I would like to know from your experience, if I'm having problems on VPN connections or failover configuration.

  Please, do not hesitate to ask as much as necessary information. All comment and documentation will be appreciated.

  Best regards!

  It's the n: documentatio

   Does not support Active/Active or Active/Standby failover

  And on top of that, ASDM shows that "Local CA cannot be configured when failover is activated".

• PIX failover: failover cable disconnected and active the unit off

  Hi all

  We have 2 PIX 515E 6.3 (3) in the failover configuration (not stateful failover). Basically, the failover works very well. Recently, we did some testing of failover and had the following situation:

  When we force the active PIX failover cable is disconnected, the rest-aid box inactive and has not changed in the active state.

  It is the 'normal' behavior or is there something wrong?

  Thank you for your response.

  Daniel Ruch

  Daniel,

  As mentioned previously, the behavior you report is expected. If the failover cable is removed from a pair of PIX failover during the race, each PIX will maintain it's State as active it or standby PIX. Remove the failover cable in effect, disables the failover of both units to avoid having two devices moving to an active state.

  Does make sense? I'm still confused what about * why * you test this though. Is this something you think that will happen in your environment?

  Scott

• PIX 6.3 (4) failover strangeness with VLAN

  I have a failover pair 535 6.3 (4) running and have experienced strange things while he was trying to get the dynamic failover to work. We use the serial cable to failover and a GE dedicated to the traffic of State via a cable directly connected x. We have a mix of standard interfaces for non - VLAN'ed, but also a physical I / f including 10 ~ VLAN. We are well within the limits of i / f allowed on the PIX so that isn't a problem. Also the

  VLAN'ed i / f on the two firewalls connects via a 802. 1 q trunk on the same switch Procurve 9315. All the required them VLAN is configured as marked on the two ports on the switch.

  The problem we had was that all as-based interfaces on the VLAN and physics i / f associated with these virtual local networks were perpetually in the State (pending) and we had no stats in the status section of the command 'show fail', which implies to me that stateful failover did not in fact. Failover works and traffic passes regardless of which firewall is enabled.

  Based on things I've read that I concluded that the problem is probably that 'Hello' messages were not being seen on each VLAN. So I did a bunch of capture on the VLAN different i / f of the PIX expecting to see outgoing Hello in the local unit, but saw nothing. Then I had a thought that maybe they were sent out without a label on the physical I / f, so I made a capture on it and also got nothing else than to the Hellos coming out to the physical interface.

  What we did that fixed it was to add the VLAN physics to the list of allowed VLANS labeled on firewall connected switch ports. As if by magic the physics I / f to the Normal State, as did all the local VIRTUAL network interfaces, and we started to get statistics on the State of the output of the show fail command section.

  And yet a capture on any of the interfaces VLAN does still not show the Hellos, and a capture on the physical displays now the bidirectional Hello for the physical LAN. Weird.

  So my questions are:

  1 > why the VLAN interfaces are dependent on their physical I / f for failover. I was told that you need not have any IP or configured for physical integrity, nameif I / f it's just must be enabled for the VLAN I / fs to work.

  2 > how are the VLAN I / f passing Hellos to the other.

  I can include my config if that helps.

  Peter

  Peter,

  (1) why is a good question. AFAIK that is according to the doc (same link below)

  "When you set up failover for an interface VLAN, Hello packets are sent through the physical interface, so the physical interface must be configured with an ip address."

  (2) I don't think that they are:

  One of the guides

  http://www.Cisco.com/en/us/products/sw/secursw/ps2120/products_configuration_guide_chapter09186a0080172786.html

  "Note that failover is supported with VIRTUAL local network interfaces. But the failover LAN interface command does not support VIRTUAL local network interfaces or failover link commands. "

  So basically it looks like helo packets are sent only on physical interfaces (dumped on any vlan you put them) and the VLAN will be "failover" If the pix, but if you had a failure in one vlan particular the pix would not notice it until the vlan the physical interface has been awarded to failed.

  Of course, it works in the equivalent level of the FWSM code - but FWSM never had the physical interfaces.

  The train 7.x supports subinterfaces, obviously.

  -Jason

  Please rate this message if it helps!

• configured with failover capability n/a

  Hello

  My configuration of cluster HA doesn't seem to work.

  The virtual machine failover does not work.

  I noticed "configured failover ability n/a" in the summary of the cluster tab.

  I was looking on the internet but can't find a solution.

  Can someone help me please.

  I'm test for my boss.

  We are currently working with linux (drbd and KVM) the virtualisize.

  But we want to implement a better solution.

  Thank you!

  John L.

  Hello

  I found the solution.

  I cut all the CD-ROMs of the virtual machines.

  And now, he is able to fialover!

  Thanks for the people who have tried to help him!