Strengthen replication integration coverage

import os

import pytest

from harness.logger import logger

import os

@pytest.mark.parametrize("client_fixture", ["gkfs_client"])

def test_replication_block_usage(test_workspace, gkfs_daemon, client_fixture, request):

    # Get the appropriate client fixture

    client = request.getfixturevalue(client_fixture)

    # Enable replication: 1 replica means 2 copies total (primary + 1 replica)

    # Client.run() starts a fresh gkfs.io process for every command and merges

    # this environment into the process environment before preload startup.

    # Setting it after fixture construction is therefore intentional.

    client._env['LIBGKFS_NUM_REPL'] = '1'

    # Verify initial state

    # Write ample data to ensure we consume multiple chunks

    # 1MB write, chunk size 512KB -> 2 chunks.

    # With replication=1, we expect 4 chunks used.

    chunk_write_size = 100 * 1024 

    chunk_write_size = 128 * 1024

    total_write_len = 1 * 1024 * 1024

    ret = client.open(file_path, os.O_CREAT | os.O_WRONLY)

    buf = b'R' * chunk_write_size

    written = 0

    while written < total_write_len:

        ret = client.write(file_path, buf, chunk_write_size, 1) # 1 for append

        ret = client.write(file_path, buf, chunk_write_size, 1)  # O_APPEND

        assert ret.retval == chunk_write_size

        written += chunk_write_size

    ret = client.stat(file_path)

    assert ret.retval == 0

    assert ret.statbuf.st_size == total_write_len

    ret = client.read(file_path, total_write_len)

    assert ret.retval == total_write_len

    assert ret.buf == b'R' * total_write_len

    # Verify updated state

    ret = client.statfs(gkfs_daemon.mountdir)

    assert ret.retval == 0

    consumed_blocks = initial_free - ret.statfsbuf.f_bfree

    expected_chunks_primary = total_write_len // chunk_size

    # We expect roughly double usage

    expected_chunks_total = expected_chunks_primary * 2

    logger.info(f"Consumed blocks: {consumed_blocks}, Expected approx: {expected_chunks_total}")

    # Allow for some variance due to block alignment etc, but it should be significantly more than primary only

    # Disabled as parallelism may affect the value

    # assert consumed_blocks >= expected_chunks_total

    logger.info(

        "Consumed filesystem blocks: %s (filesystem block size: %s)",

        consumed_blocks,

        chunk_size,

    )

    # Clean up

    client.unlink(file_path)