[BUG] Replication incomplete when disk space is low

[keithchew]

Testing on v6.3.4 (with flash), but this bug should be present in previous versions as well.

Steps:

• master and slave both in sync
• ensure disk space on master is close to full, and not enough to create dump.rdb
• restart master
• upon restart, it will trigger full sync replication
• slave does not get the full dataset from master, causing havoc at the app level.

On my local I have added an abort flag:

        bool fContinue = fn(it->key().data()+HASHSLOT_PREFIX_BYTES, it->key().size()-HASHSLOT_PREFIX_BYTES, it->value().data(), it->value().size());
        if (!fContinue) {
            aborted = true;
            break;
        }
...
    if (aborted) {
        printf("WARNING: rocksdb enumerate signaled to abort, m_count [%lu] count [%lu]\n", m_count, count);
        return false;
    }

This has fixed the issue, logs below:

7:71:M 2 Apr 2024 19:51:43 WARNING: rocksdb enumerate signaled to abort, m_count [5069771] count [4822420]
7:71:M 2 Apr 2024 19:51:43 * rdbSaveRio Failed to save
7:71:M 2 Apr 2024 19:51:43 # Write error saving DB on disk: No space left on device