Attention is currently required from: laforge, neels.

View Change

2 comments:

• Patchset:

  • Patch Set #3:

    Several pointers below:

    • I think that definetly we want to merge all this done in a separate thread from first step. I don't think it even makes sense to have it separated in a first step where everything happens in the same thread, since it's broken and distracts the whole thing.

    We don't have multi-threaded rate counters so far anywhere.

    • We do, in osmo-trx. In there since it's heavily multithreaded we do things a bit adhoc with separate structures and then copying them.

    The other idea that does not need libosmocore changes is this:
    have a separate "volatile" counter storage in the hnb_persistent struct;
    in the nft thread, do all parsing and all hnb lookups,
    and place parsed counters in that separate counter storage per hnbp.
    inter-thread signal (it_q is nice because it integrates in the select())
    to tell the main thread to shovel all hnb volatile counters into the rate_ctr;
    those volatile counters need a mutex (one global one, or maybe one per hnb).

    That's similar to how osmo-trx does it, and I think it's how it should be done.
    All other ways imho are calling for inconsistencies and stuff ending up broken.

    So my opinion is: I'm waiting to see a patch like the above described which I can review.

• Patchset:

  • Patch Set #17:

    ree

To view, visit change 36385. To unsubscribe, or for help writing mail filters, visit settings.