Due to the limitation of the underlying storage engine, each key-value entry (one row) in TiDB should be no more than 6MB. You can adjust the txn-entry-size-limit configuration value up to 120MB.
Distributed transactions need two-phase commit and the bottom layer performs the Raft replication. If a transaction is very large, the commit process would be quite slow and the write conflict is more likely to occur. Moreover, the rollback of a failed transaction leads to an unnecessary performance penalty. To avoid these problems, we limit the total size of key-value entries to no more than 100MB in a transaction by default. If you need larger transactions, modify the value of txn-total-size-limit in the TiDB configuration file. The maximum value of this configuration item is up to 10G. The actual limitation is also affected by the physical memory of the machine.
There are similar limits on Google Cloud Spanner.