Skip to content

test_volume_io hang after several rounds of IO load and restarts (graceful) #151

@yamingk

Description

@yamingk

After several rounds of restarts, the HomeBlks/HomeStore will hang during IO load. This seems to be a cause of we can not sucessfully get a chunk select_chunk() during IO path, and it kept trying and hangs there.
If this is because of volume full, HomeBlks should be able to break out of the while loop and returning error because of no available_chunks. However this is unlikely the case, as from the logs it did 4 rounds of restarts and each round only do 300 IOs with mixed size.

homestore::cshared< Chunk > VolumeChunkSelector::select_chunk(homestore::blk_count_t nblks,
                                                              const homestore::blk_alloc_hints& hints) {
...
    do {
...
        LOGT("Waiting to allocate more chunks active={} total={}", volc->num_active_chunks.load(),
             volc->max_num_chunks);
        LOGT("{}", dump_chunks());
        std::this_thread::sleep_for(std::chrono::milliseconds(100)); // <<<< we are looping here
    } while (true);
...

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions