[00:25:29] *** Joins: gila (~gila@5ED4FE92.cm-7-5d.dynamic.ziggo.nl) [02:53:46] *** dstojacx is now known as darsto [03:14:51] hi all, Darek Stojaczyk here [06:54:49] hi darsto - glad to see you on IRC :) [07:55:58] *** Joins: lhodev (~Adium@inet-hqmc08-o.oracle.com) [08:39:32] *** Quits: lhodev (~Adium@inet-hqmc08-o.oracle.com) (Quit: Leaving.) [08:57:12] *** Quits: shiyer (~shiyer@143.166.116.70) (Ping timeout: 246 seconds) [09:13:38] welcome to the party :) [09:41:14] I made a trello board for all of the blobstore/blobfs/rocksdb work https://trello.com/b/GM6v8FG6/blobstore [10:50:06] *** Quits: gila (~gila@5ED4FE92.cm-7-5d.dynamic.ziggo.nl) (Quit: My Mac Pro has gone to sleep. ZZZzzz…) [10:52:21] *** Joins: gila (~gila@5ED4FE92.cm-7-5d.dynamic.ziggo.nl) [12:39:24] *** Joins: lhodev (~Adium@inet-hqmc08-o.oracle.com) [12:41:00] jimharris: if the first patch of darsto's virtio series looks good, I can rebase it and merge it: https://review.gerrithub.io/#/c/375005/ [12:41:21] I think we should discuss whether we actually want to go back to async module init, though - does it actually buy us anything? [12:53:27] you'd prefer to just poll? [12:54:00] I guess we can benchmark it - surely it's way less time than it takes to wait for an NVMe SSD to initialize [12:56:03] yeah, I don't know that it helps us at all to do async init, unless we can actually overlap module inits [12:56:10] but that's not what it did before, right? we just did one at a time [12:57:01] before it was just one at a time but his implementation now expands on what I did for vbdev examination to allow overlapping inits [12:57:24] hmm, that might be more interesting [13:04:46] *** Parts: lhodev (~Adium@inet-hqmc08-o.oracle.com) () [13:34:55] what if instead of requiring each module to call spdk_bdev_module_init_done(), we instead allowed modules to "opt-in" to an asynchronous init? [13:50:46] that would be better - my main concern with the current approach is that all modules have to deal with it [13:59:42] we could either have a function that virtio-scsi has to call to mark itself as async, or we have module init return a special value to denote async init [14:02:00] return value doesn't really work - some modules return errors with positive return codes, others with negative [14:12:32] calling something sounds reasonable - can it call a "start async init" request from within its normal module init callback? [14:12:56] how about adding SPDK_BDEV_MODULE_ASYNC_INIT(name) [14:13:21] you just add that after SPDK_BDEV_MODULE_REGISTER in whatever module wants it - it just initializes the actions_in_progress to 1 [14:13:45] sounds fine [14:30:40] does the lvol patch depend on anything else going in first, or is it ready to be reviewed? [16:27:53] *** Quits: gila (~gila@5ED4FE92.cm-7-5d.dynamic.ziggo.nl) (Quit: My Mac Pro has gone to sleep. ZZZzzz…) [17:11:48] bwalker, drv: https://review.gerrithub.io/#/c/375651/ [17:11:56] lvol patch can be reviewed now [18:42:55] *** Quits: konv81 (cf8c2b51@gateway/web/freenode/ip.207.140.43.81) (Ping timeout: 260 seconds)