From: John W. <jl...@sc...> - 2011-07-21 20:36:06
Attachments:
hostd.part.log
|
I have the basics for vmtools ported for OpenServer 6. The RPC loop is running and the hostd.log file on the ESX host reports the current heartbeat status as green. But, the vSphere Client Summary tab for the running VM indicates: "VMware Tools: Not Installed" And it has consistently been displayed that way except for about 1/2 hour of testing during which the status was displayed as "Unmanaged" while the RPC loop was running, and "Not Running" when vmtoolsd was killed. The "Not Installed" status has returned, but that got me digging a little bit. Managed Object Browser: =========================================== The Managed Object Browser (MOB) of the GuestInfo for my running OSR6 VM shows: toolsRuningStatus string "guestToolsRunning" toolsStatus VirtualMachineToolsStatus "toolsOK" toolsVersion string "2147483647" (32 bit MAXINT - 0x7fffffff) - probably the TOOLS_VERSION_UNMANAGED toolsVersionStatus string "guestToolsNotInstalled" Now, from the top three items, I would have expected the "VMware Tools:" status to at least indicate "Unmanaged". Clearly the last item "toolsVersionStatus" is being set incorrectly (?) and that is apparently what results in the VMware Tools: Not Installed being displayed in the vSphere Client Summary tab for a running VM. I have confirmed that services/vmtoolsd/toolsRpstatusc.c is sending the "tools.set.version 2147483647" ESX 4.0 (build-398348): /var/log/vmware/hostd.log =================================================== Looking in the ESX 4.0 host /var/log/vmware/hostd.log I see: - the tools status is changed to "running" - the tools version changing from 0 (default) to TOOLS_VERSION_MANAGED - and then a struggle between "Unexpected toolsVersionStatus notAvailable - using offline status guestToolsNotInstalled" and "changing status to running" - repeating within msecs - not on to 30 sec vmtools polling interval - rapidly filling /var/log/vmware/hostd.log - section of log attached (hope the attachment is usable) /vmfs/volumes/*/<VM>/vmware.log file contains: =============================================== Jul 21 14:24:12.677: vcpu-1| GuestRpc: Channel 0, guest application toolbox. Jul 21 14:24:12.677: vcpu-1| Vix: [104286 mainDispatch.c:3169]: VMAutomationReportPowerStateChange: Reporting power state change (opcode=2, err=0). Jul 21 14:24:12.696: vcpu-1| TOOLS ToolsCapabilityGuestConfDirectory received /etc/vmware-tools Jul 21 14:24:12.701: vcpu-1| TOOLS setting legacy tools version to '2147483647', manifest status is 4 Jul 21 14:24:12.702: vcpu-1| DISKLIB-DDB : "toolsVersion" = "2147483647" (was "0") Jul 21 14:24:12.732: vcpu-1| VMXVmdb_SetToolsVersionState: status value set to 'notAvailable' Jul 21 14:24:12.732: vcpu-1| TOOLS unified loop capability requested by 'toolbox'; now sending options via TCLO Jul 21 14:24:12.734: vcpu-1| TOOLS sending 'OS_PowerOn' (3) state change request Jul 21 14:24:12.735: vcpu-1| Guest: toolbox: Version: build-402641 Jul 21 14:24:12.815: vcpu-0| TOOLS state change 3 returned status 1 The question is what am I forgetting to set? Thanks. -- John Wolfe UnXis, Inc. |
From: Marcelo V. <mv...@vm...> - 2011-07-21 20:47:10
|
Hi John, On 07/21/2011 01:32 PM, John Wolfe wrote: > Jul 21 14:24:12.732: vcpu-1| VMXVmdb_SetToolsVersionState: status value set to > 'notAvailable' I'm pretty sure this is a bug in VMware's host code, so there's nothing you can do from the Tools side. Someone else reported a similar issue here before, and we've already fixed the host code in the vSphere 5.0 release. (I don't think the fix was backported to any maintenance release of previous vSphere versions.) -- - Marcelo |
From: John W. <jl...@sc...> - 2011-07-21 22:05:12
|
Marcelo, Petr, Is the "bug" in vSphere 4.x flaky enough for the tools status to occasionally be reported as "unmanaged" for unsupported guests? - yesterday my OSR6 VM was "Unmanaged" for several reboots and restarts of the vmtoolsd - today, following a power-hit to our Intel Modular Servers, an OSR507 VM was rebooted and is reporting a VMtools status of "Unmanaged" - through several shutdown and power-up cycles. Previously the status was being reported as "Not Installed". Is there anything that I can do to reduce the number of tools running, not running, version not available, not installed messages that are being dropped into /var/log/vmware/hostd.log? -- John Marcelo Vanzin wrote: > Hi John, > > On 07/21/2011 01:32 PM, John Wolfe wrote: > >> Jul 21 14:24:12.732: vcpu-1| VMXVmdb_SetToolsVersionState: status value set to >> 'notAvailable' >> > > I'm pretty sure this is a bug in VMware's host code, so there's nothing you can > do from the Tools side. Someone else reported a similar issue here before, and > we've already fixed the host code in the vSphere 5.0 release. (I don't think the > fix was backported to any maintenance release of previous vSphere versions.) > > |
From: Marcelo V. <mv...@vm...> - 2011-07-21 22:53:28
|
Hi John, On 07/21/2011 03:01 PM, John Wolfe wrote: > Is the "bug" in vSphere 4.x flaky enough for the tools status > to occasionally be reported as "unmanaged" for unsupported guests? The flakiness of the status does look weird; but looking at the fix for the bug, it does seem like it would properly handle this case. The bug was that the status could be set to "not installed" if the configured guest OS did not have a VMware tools package (which would be the case for "other" or "sco" guest OS types). I don't know if "could" there means that the status would always be set to "not installed". But with the fix, if the Tools version is "unmanaged", that's always properly reported, regardless of the OS type. > Is there anything that I can do to reduce the number of tools > running, not running, version not available, not installed > messages that are being dropped into /var/log/vmware/hostd.log? The workaround that the user who first complained used was to set up the VM as "linux" or "linux 64-bit". That would work around this particular bug, since we do have a Tools package for Linux. I'm not sure whether ESX 4.1 has a specific guest os type for SCO. I think we have some internal tweaks for SCO regarding time keeping, that are only applied if you choose the appropriate guest OS type. So changing it to Linux might have some adverse side-effects. -- - Marcelo |
From: John W. <jl...@sc...> - 2011-07-26 15:47:36
|
Marcelo Vanzin wrote: > I don't know if "could" there means that the status would always be set to "not > installed". Could vs. would seems to suggest some variability in observed behavior. We can assure OSR5, OSR6 or probably UW714 VM users on ESX 4.x hosts that the display of the IP(s) and system name indicate that open-vm-tools are installed and running on their VM. That should satisfy everyone. > The workaround that the user who first complained used was to set up the VM as > "linux" or "linux 64-bit". That would work around this particular bug, since we > do have a Tools package for Linux. > I think that it would be preferable to display the correct Guest OS type. (See below) > I'm not sure whether ESX 4.1 has a specific guest os type for SCO. I think we > have some internal tweaks for SCO regarding time keeping, that are only applied > if you choose the appropriate guest OS type. So changing it to Linux might have > some adverse side-effects. > Our current OSR 5.0.5 VA released was generated on WS for Windows 6.5.x as: Other: Other-32 (No SCO guest type available then) VM version: 4 open-vm-tools: 2010.03.20-243334 INFO_OS_NAME = null string. and that guest OS type is carried through the import of the VA and running of the VM. Our OSR6 efforts are starting with generation of the VM on ESX 4.0 which offers guest OS types for "SCO OpenServer 5" and "SCO UnixWare 7", but nothing for OpenServer 6. Other: Other-32 VM Version: 7 open-vm-tools: 2011.04.25-402641 (port starting point) INFO_OS_NAME = "openServer6" I just invented an OS_NAME string which triggered a warning in the hostd.log file about an unrecognized/unsupported guest OS. A search through the ESX host files revealed Guest OS types for "openServer5", "openServer6" and "unixWare7". When open-vm-tools was modified to provide INFO_OS_NAME as "openServer6", the vSphere Client will switch from "Other-32" to "SCO OpenServer 6" on a running VM. The next release of open-vm-tools for OpenServer 5.0.7 will by corrected to provide "openServer5". -- John Wolfe |
From: Marcelo V. <mv...@vm...> - 2011-08-01 21:49:48
|
Hi John, On 07/26/2011 08:43 AM, John Wolfe wrote: > Our OSR6 efforts are starting with generation of the VM on > ESX 4.0 which offers guest OS types for "SCO OpenServer 5" and > "SCO UnixWare 7", but nothing for OpenServer 6. The identifier provided by the guest os is, mostly, used for the UI only. The other uses are for features that wouldn't support SCO anyway, so having it not completely accurate wouldn't really be a problem aside from the cosmetic issue. The warning you saw in hostd's logs is because it expects the guest to provide exactly the same string it has in its internal table; looking at the ESX 4.0 sources it seems there are three flavors of SCO: openserver5, openserver6 and unixware7. I don't know for sure but my guess is that they are case-sensitive. They're also not marked as fully supported, although I don't know if that flag affects anything in the UI (e.g. whether to show the OS when setting up a new VM). But none of this fixes the original bug that was fixed on ESX 5.0, where the Tools status would be incorrectly reported for OSes that don't have official VMware Tools packages. So I can't really tell you what the exact behavior there would be, because I don't know. -- - Marcelo |