Change in ovirt-engine[master]: bootstrap: do not get unique id at canDoAction

yzaslavs at redhat.com yzaslavs at redhat.com
Mon Nov 19 11:58:37 UTC 2012


Yair Zaslavsky has submitted this change and it was merged.

Change subject: bootstrap: do not get unique id at canDoAction
......................................................................


bootstrap: do not get unique id at canDoAction

CURRENT IMPLEMENTATION

engine has duplicate complex vdsm logic to generate the vdsm id.

    @Reloadable
    @TypeConverterAttribute(String.class)
    @DefaultValueAttribute(
        "IDFILE=/etc/vdsm/vdsm.id; " +
        "if [ -r \"${IDFILE}\" ]; then " +
            "cat \"${IDFILE}\"; " +
        "else " +
            "UUID=\"$(" +
                "dmidecode -s system-uuid 2> /dev/null | " +
                "sed -e 's/.*Not.*//' " +
            ")\"; " +
            "if [ -z \"${UUID}\" ]; then " +
                "UUID=\"$(uuidgen 2> /dev/null)\" && " +
                "mkdir -p \"$(dirname \"${IDFILE}\")\" && " +
                "echo \"${UUID}\" > \"${IDFILE}\" && " +
                "chmod 0644 \"${IDFILE}\"; " +
            "fi; " +
            "[ -n \"${UUID}\" ] && echo \"${UUID}\"; " +
        "fi"
    )
    BootstrapNodeIDCommand(372),

The command is executed synchronously (UI wise) when host is added.

ASSUMPTIONS OF CURRENT IMPLEMENTATION

 1. dmidecode exists out of the box in any distribution.

 ---> WRONG: fedora 17 has no, and other distributions may also lack.

 2. host id is only dmidecode output.

 ---> WRONG: over time we saw that we need extra logic to keep the id
      sane, especially when the hardware id does not exist or
      malformed.

 3. dmidecode utility is used for host id

 ---> WRONG: there are plans to make it more secure/robust using TPM,
      which requires software at host to generate.

PROBLEMS IN CURRENT IMPLEMENTATION

 1. if dmidecode utility is missing we cannot acquire host id before
    performing bootstrap. The whole idea of bootstrap process is to take
    vanilla distribution and install vdsm. dmidecode is missing from
    vanilla, hence cannot be executed before bootstrap.

 2. the logic of generating host id exists both in engine and vdsm, both
    implementations need to synced, and kept synced between versions,
    which in practice cannot be achieved. As there is too much static
    noise (distribute on different channels, not be able to update all
    IT components, etc...).

 3. If host id generation method is changed, the engine implementation
    should be changed as well, while engine should not really care how
    vdsm maintain its identity.

NEW IMPLEMENTATION

Acquire vdsm id during bootstrap process, at the earliest. When all
dependencies are available.

If vdsm id is duplicate:

 1. post an error to host event log.

 2. set the state of the host to "install fail".

USER VISIBLE CHANGES

The following existing scenario:

 1. user has host xxx.com with ip 1.1.1.1, he added this host using
    xxx.com as host name.
 2. user add new host, at the host field he *BY MISTAKE* writes 1.1.1.1.
 3. the user is blocked from proceeding because of duplicate uuid between
    xxx.com(existing) and 1.1.1.1(new).
 4. the confused user fixes the  host field to 1.1.1.2 and proceed.

Behaves as:

 1. user has host xxx.com with ip 1.1.1.1, he added this host using
    xxx.com as host name.
 2. user add new host, at the host field he *BY MISTAKE* writes 1.1.1.1.
 3. installation starts, and ends up with installation failed status on
    the new host.
 4. user sees the error message, and remove the host added by mistake.

REASONING

The above sequence of adding the same host with different name/address is not
common or frequent, and does not justify the need to duplicate vdsm logic into
engine, nor find a solution for distributions that locks the dmidecode utility
at their base system layout, nor to lock our-self to dmidecode utility.

Bug-Url: https://bugzilla.redhat.com/show_bug.cgi?id=875527
Change-Id: I0263dbae34aaa02c126c5ed1dc52a84f4f5e77f8
Signed-off-by: Alon Bar-Lev <alonbl at redhat.com>
---
M backend/manager/modules/bll/src/main/java/org/ovirt/engine/core/bll/AddVdsCommand.java
D backend/manager/modules/bll/src/main/java/org/ovirt/engine/core/bll/VdsInstallHelper.java
M backend/manager/modules/bll/src/main/java/org/ovirt/engine/core/bll/VdsInstaller.java
M backend/manager/modules/bll/src/test/java/org/ovirt/engine/core/itests/BasicTestSetup.java
4 files changed, 64 insertions(+), 180 deletions(-)

Approvals:
  Alon Bar-Lev: Verified
  Yair Zaslavsky: Looks good to me, approved


--
To view, visit http://gerrit.ovirt.org/9159
To unsubscribe, visit http://gerrit.ovirt.org/settings

Gerrit-MessageType: merged
Gerrit-Change-Id: I0263dbae34aaa02c126c5ed1dc52a84f4f5e77f8
Gerrit-PatchSet: 3
Gerrit-Project: ovirt-engine
Gerrit-Branch: master
Gerrit-Owner: Alon Bar-Lev <alonbl at redhat.com>
Gerrit-Reviewer: Allon Mureinik <amureini at redhat.com>
Gerrit-Reviewer: Alon Bar-Lev <alonbl at redhat.com>
Gerrit-Reviewer: Barak Azulay <bazulay at redhat.com>
Gerrit-Reviewer: Doron Fediuck <dfediuck at redhat.com>
Gerrit-Reviewer: Douglas Schilling Landgraf <dougsland at redhat.com>
Gerrit-Reviewer: Juan Hernandez <juan.hernandez at redhat.com>
Gerrit-Reviewer: Yair Zaslavsky <yzaslavs at redhat.com>



More information about the Engine-commits mailing list