Handling configuration drift in backup storage systems

1. A computer-implemented method of handling configuration drift in a data storage system having a plurality of storage nodes, comprising:

defining a golden configuration dataset for the data storage system;

obtaining a current configuration dataset of each storage node of the plurality of storage nodes, each of the golden and configuration datasets comprising a plurality of sentences defining a node configuration parameter;

determining a distance between each sentence of the golden configuration dataset with each other sentence of the current configuration datasets for each of the plurality of storage nodes by creating a probability distribution of all sentence embeddings in the n-dimensional space using an unsupervised clustering technique, generating probability distributions for each of the golden configuration dataset and current configuration datasets, and creating, for each probability distribution, a distance matrix representing a pair-wise distance between each point and every other point in the probability distribution;

ranking each node based on a distance of its sentences with the golden configuration dataset;

performing an action on a corresponding node based on its respective ranking;

embedding every sentence as a vector in n-dimensional space wherein every sentence has a point value representing its position relative to all other sentences, and further wherein a distance corresponds to a measure of closeness between two sentences.

2. The method of claim 1 further comprising using a Bidirectional Encoder Representations from Transformers (BERT) process to create the probability distribution of all the sentence embeddings.

3. The method of claim 1 further comprising applying an optimal transport process to each pair of probability distributions to identify contextual configuration drift therebetween.

4. The method of claim 3 further comprising applying a Wasserstein distance technique to the optimal transport process to compensate for any misalignment of axes in the n-dimensional space.

5. The method of claim 4 further comprising:

normalizing each distance matrix for a current configuration dataset of a respective node to the golden configuration dataset to a scale from zero to one; and

generating a score for each node along the scale based on a respective distance matrix.

6. The method of claim 5 further comprising:

defining one or more threshold values defining a relative severity of configuration drift along the scale;

placing the score for each node along the scale relative to the threshold values; and

triggering an automated response for the node based on a respective score for the node along the scale.

7. The method of claim 6 wherein the automated response comprises one of:

take no action for scores below a first threshold value;

generate an alert for scores between a first threshold value and a second threshold value; and

push the golden configuration onto the current configuration for scores above the second threshold value.