HPE Ezmeral Data Fabric (MapR) Installation Failed with rpcbind disabled

[SunnyC]

Background :

We are planning for a server update from MapR 6.2.0 to 7.7.0. Due to the change from a non-secure cluster to a secured cluster, we are doing fresh installation. We are using RHEL 8.6 and applied hardening based on CIS Red Hat Enterprise Linux 8 Benchmark - Level 1.

Problem:

We have masked rpcbind and rpcbind.socket in our environment:

During Installer installation, it failed with below error:

status: Unable to enable service rpcbind: Failed to enable unit: Unit file /etc/systemd/system/rpcbind.service is masked.

Is there anyway to disable nfs setup during installation so rpcbind won't be needed? 

[support_s]

System recommended content:

1. HPE Ezmeral Data Fabric – Customer-Managed 7.6.1 Documentation | Before You Start Using Data Fabric NFS

2. HPE Ezmeral Data Fabric – Customer-Managed 7.6.1 Documentation | Starting the mapr-loopbacknfs Service to Access a Cluster

Please click on "Thumbs Up/Kudo" icon to give a "Kudo".

Thank you for being a HPE valuable community member.

[Dave Olker]

To be clear, how are you performing the installation?  Are you using the installer or doing a manual installation?  How many nodes are in the cluster? 

[SunnyC]

@Dave Olker I have tested different kinds of installation for this issue.

The provided error occurs during both single node and 4-node cluster fresh installation with installer.

Below are the services I select during the installation (please ignore the version number):

[Dave Olker]

I was able to reproduce this in my lab.  Even when selecting to not install any NFS v3 or v4 and no Ecosystem components the installer is still checking for rpcbind:

2024-05-27 08:25:55,298 callbacks 415 [INFO]: Checking if rpcbind is started(service {
"enabled": "yes",
"name": "rpcbind",
"state": "started"
}) Unable to enable service rpcbind: Failed to enable unit: Unit file /etc/systemd/system/rpcbind.service is masked.

This is a design issue with the installer so it will take time to change the installer logic.  I will open an engineering ticket to report this issue.   My suggestion at this point would be to manually install the cluster without the Installer using the instructions here: 

https://docs.ezmeral.hpe.com/datafabric-customer-managed/77/AdvancedInstallation/InstallingMapRSoftware.html

[SunnyC]

Sorry for the late reply.

I was setting up a testing environment and was still facing some error during manual installation.

The new environment is a single node Core 7.6.1 on RHEL 8.6 with openjdk 11.

Below core packages are installed:

Below is the config cmd:

# /opt/mapr/server/configure.sh -secure -genkeys -Z <hostname> -C <hostname> -N my.cluster.com -F /tmp/disklist

The config finished without error but I can find some new error in gateway and cldb logs.

gatewayinit log is filled with below line and the log growth to 25G in less than 1 hour.

gatewayinit.log:2024-05-29 06:52:50,0864 : 535 Failed to connect to IPv4 [127.0.0.1]:0 error: 111 sock: 167

cldb log also shows some java error as below:

I tried the same installation approach after I unmask rpcbind and rpcbind.socket and reboot server and all these error are gone.

[Dave Olker]

Sounds like there are dependencies on rpcbind even in manual deployment.  The error you're seeing comes from our rpcbinding.cc code, and errno 111 is "ECONNREFUSED 111 Connection refused".  This means there are hooks in the code to do an RPC binding, which I would expect requires contacting the rpcbind daemon, and since it is not running an errno 111 would make sense.

I can add this information to the engineering ticket I opened previously.  However, the fact that our code is explicitly calling rpcbind outside of NFS tells me there may not be a good fix for this masking of rpcbind issue.

[SunnyC]

Thank you for the update.

We will continue our development with Core Version 7.1.0 at this moment.

Is there any way we can get an official statement stating that rpcbind service is an OS dependency for certain Core Service?

[Dave Olker]

Not at this time.  I've got engineering looking into this to see if this dependency on rpcbind can be removed from the cluster installation process.  I'll post when I have an update.