"Designing a Dropbox-like File Storage Service - A Performance and Cost study"
If you are confused about stuff in this design and contradictions, do yourself a favour and read this original paper. Educative did a poor job of “even” copying from the original paper.
Still the logical diagram in the paper is a little wrong as client should contact the Metadata server directly instead of going through the synchronization service (which is a complete contradiction with the written paragraphs)
Once you are done reading the paper, watch the stanford video on youtube for " How We’ve Scaled Dropbox" by dropbox architect.
These 2 resources will give you better idea.
And for god sake, do not use “DynamoDB” for metadata server as it only provides eventual consistency. We need here strong consistency and availability so something like Google Spanner will be a way better choice than this.
Which again Educative copied word-by-word from the above paper that also mentions to use DynamoDB.