Failed to start controller : can t find valid default client configuration template file or file not json about beegfs-csi-driver

[seb-835]

i search and got more details on my case,

in the sts deploiement file, the controller is set to run on master/control place node.
On my infrastructure, only nodes have infiniband and beegfs tools and setting, so that' s why it failed looking for /etc/beegfs/beegfs-client.conf file on master nodes.

Anyway, i copy the file to the master nodes, and now got :

"failed to handle configuration file: likely missing quotes around an integer or boolean beegfsClientConf value: error unmarshaling JSON: while decoding JSON: json: cannot unmarshal number into Go struct field BeegfsConfig.config.beegfsClientConf of type string" "fullError"="error unmarshaling JSON: while decoding JSON: json: cannot unmarshal number into Go struct field BeegfsConfig.config.beegfsClientConf of type string\nlikely missing quotes around an integer or boolean beegfsClientConf value\ngithub.com/netapp/beegfs-csi-driver/pkg/beegfs.parseConfigFromFile

That's true the file is a not json file.
Can you share an example of your /etc/beegfs/beegfs-client.conf for me to compare with mine and then i can adapt it on my controlplane nodes

[ejweber]

Hello @seb-835. There are likely a couple of things at play here:

1. You mentioned that master nodes in your infrastructure are incapable of running (or do not run) BeeGFS services. While the initial error you get when trying to run the BeeGFS CSI driver on nodes like this relates to a missing beegfs-client.conf file, simply copying the file to these nodes is not a fix. The driver will eventually fail for other reasons (as it must be able to use BeeGFS related modules and programs on the node where it runs). The correct approach is to patch the deployment manifests so that the driver is never scheduled to run on incapable nodes. See https://github.com/NetApp/beegfs-csi-driver/blob/master/docs/deployment.md#deployment-to-kubernetes-clusters-with-mixed-nodes for detailed instructions. Feel free to reach out with specific questions if this does not go well.
2. The unmarshalling error you are seeing is not occurring when the driver is reading /etc/beegfs/beegfs-client.conf (this file should generally remain in its default state). It is occurring when the driver reads in your csi-beegfs-config ConfigMap. You have included a large number of parameters here, many of which are simply the defaults. It is only necessary to specify parameters in this file if/when they need to be different than the defaults specified in /etc/beegfs/beegfs-client.conf. However, if you do specify parameters here, booleans and integers must be quoted as strings. I recommend you remove any parameters from the ConfigMap you don't need and fix any booleans or integers you do need.
3. Note that it is incorrect to specify sysMgmtdHost in the beegfsClientConf section of the ConfigMap (the driver will log a warning). sysMgmtdHost is specified individually for each PersistentVolume (and set automatically on provisioned PersistentVolmes according to the StorageClass parameters).

[seb-835]

Hi @ejweber ,
i am very grateful for this long and detailed answer, many thanks, this resolve my trouble !!!

[ejweber]

Glad to hear the issue is resolved!