Cathy Zhou wrote: > Tony Nguyen wrote: >> Cathy Zhou wrote: >>> >>>>> We have a few questions for the SMF folks, which is why they're >>>>> CC-d this mail. Is our use of SMF sounds reasonable? Do we need >>>>> to make the linkmgmtd service depend on anything? We didn't know >>>>> how SMF persists configuration information, and we didn't know if >>>>> we needed to include a dependency here to make sure we can access >>>>> the configuration data. >>>> >>>> It'd be a good idea to have a dependency on manifest-import. This >>>> service generates the repository content on first reboot following >>>> the installation and gets updated on upgrades. Services attempts to >>>> access repository before manifest-import completes may get errors >>>> because properties are not populated or old property values. The >>>> drawback of having this dependency is the service may come up later. >>>> >>> Hmm, I saw that manifest-import depends on device/local and >>> filesystem/usr. That will cause the dependency recursion as the >>> device/local needs to have the dependency on the linkname service, >>> in order to get the link name when the network devices attaches. >>> >>> - Cathy >> Cathy, >> >> That's not what I'm seeing on b57 or S10 01/06. Having dependency on >> manifest-import allows updates to your service manfest(on upgrades) >> to be imported into the repository before the service starts. If this >> dependency doesn't exist, your service may get started with the old >> service definition. >> >> -tony >> >> bash-3.00$ svcs -l manifest-import >> fmri svc:/system/manifest-import:default >> name service manifest import >> enabled true >> state online >> next_state none >> state_time Mon Feb 12 17:21:09 2007 >> logfile /var/svc/log/system-manifest-import:default.log >> restarter svc:/system/svc/restarter:default >> dependency require_all/none svc:/system/filesystem/minimal (online) >> bash-3.00$ > > This shows manifest-import depends on filesystem/minima, which depends > on device/local and filesystem/usr. > > I think linkmgmtd service must be part of seed repository, and I think > this should be sufficient so that linkmgmtd doesn't need to depend on > manifest-import. > linkmgmtd qualifies as boot-critical service and a part of the seed repository if network/physical really depends on it.
-tony
