[00:57:30] *** Quits: Shuhei (caf6fc61@gateway/web/freenode/ip.202.246.252.97) (Ping timeout: 260 seconds) [01:47:16] *** Joins: tomzawadzki (tomzawadzk@nat/intel/x-bcigpypgviscuzgl) [02:33:33] *** Joins: toka (86bfdc4a@gateway/web/freenode/ip.134.191.220.74) [05:49:44] *** Quits: toka (86bfdc4a@gateway/web/freenode/ip.134.191.220.74) (Ping timeout: 260 seconds) [06:33:15] Hi folks, could you take a look at https://review.gerrithub.io/c/399367/ ? This patch is first step for allowing deletion of snapshots that we discussed last week. This is not full implementation, only request for comments. [07:39:41] *** Joins: Shuhei (caf6fc61@gateway/web/freenode/ip.202.246.252.97) [08:02:23] Hi Tomasz and All, meeting will start soon? [08:02:33] Shuhei: yes [08:02:42] it's starting now [08:02:43] Thank you! [08:03:43] *** Joins: toka (86bfdc49@gateway/web/freenode/ip.134.191.220.73) [08:03:57] I'm connecting to the Skype but recently I replaced the PC and hence I'm not sure it is working. [08:04:32] Hi Dariusz, already started? [08:04:41] yep, we're in [08:05:19] Oh, its not working yet. thanks [08:09:52] you can try joinung via phone: +1(916)356-2663 Choose bridge 5, Conference ID: 656560618 [08:11:30] thanks [08:12:44] Actually, the Conference ID: 184891507. Sorry for confusion [08:12:59] ok thanks [08:19:00] *** Joins: johnmeneghini (~johnmeneg@pool-96-252-112-122.bstnma.fios.verizon.net) [08:19:31] Is the skype meeting supposed to be started? [08:19:56] yes, but in the invitation there was invalid conference id [08:20:03] the Conference ID: 184891507. Sorry for confusion [08:21:26] OK thanks. Can someone please email this out. [08:22:45] I've replied to your questions on the mailing list, also provided an updated link for Skype meeting [08:23:16] *** Joins: mgm (824cba0b@gateway/web/freenode/ip.130.76.186.11) [08:28:19] any advise with compiling an app with spdk and dpdk dependancies? The makefiles are a little gnarly. [08:29:24] mgm: any specific errors? [08:29:47] drv, FYI the problem with the bdev CI ASAN tests is lack of coordination on tear down of multiple split vbdevs specifically around vbdev_split_destruct(). Didn't 100% debug yet but will circle back after I figure out my other issues and put a real fix in place [08:29:54] oh, nice [08:29:59] undefined references in the linker [08:30:14] we've got scripts/pkgdep.sh scripts that should get you prepared [08:30:22] script* [08:30:38] drv, yeah, easy to see the problem and easy to hack in a fix but I'll need to spend a little more time to get something in there that isn't a band-aid [08:30:39] peluse: we should get sethhowe to look into re-enabling ASAN on the test machines - I think it's currently disabled for some reason on most/all test machines [08:30:56] drv, yeah, this is one of those reasons :) [08:31:35] mgm: have you seen http://www.spdk.io/doc/getting_started.html ? [08:32:47] johnmeneghini: were you able to join the call? [08:32:58] yea. i've built the libraries and examples for both projects. i'm just trying to create a new project from scratch. [08:33:29] ahh i get it [08:35:33] mgm, looking at the makefiles and includes for something really simple like /examples/blob/hello_blob is a good place to start [08:35:41] I'd like drv to confirm that, but you should probably start looking at examples/nvme/hello_world [08:36:10] yeah, unfortunately we don't currently have a nice concise list of linker options to use from outside of the SPDK build directory [08:36:46] k. thanks. [08:36:52] either will work... the one I mention uses the spdk event framework and bdev layers... [08:37:17] the nvme hello_world is a great example of using the nvme driver [08:37:23] there is planned work to create a pkg-config .pc file to make this easier, but that's not done yet [08:41:50] @pzedlews Skype says I'm waiting in the lobby. Do you have control over letting me in? [08:42:39] I'm seeing notifications there from time to time and I'm 'admiting' everyone [08:43:08] can you try again? [08:43:12] OK. I must be connected to the wrong meetiing. [08:43:19] I'll try again. [08:43:29] yeah. we gotta fix this [08:45:04] Is this the correct URL for the skype meeting? https://meet.intel.com/piotr.zedlewski/4BV132KR [08:45:49] nope, try this one: https://meet.intel.com/piotr.zedlewski/S3QPDH70 [08:46:37] I've just clicked Admit to someone [08:48:11] *** Quits: mgm (824cba0b@gateway/web/freenode/ip.130.76.186.11) (Quit: Page closed) [09:16:14] *** Quits: Shuhei (caf6fc61@gateway/web/freenode/ip.202.246.252.97) (Ping timeout: 260 seconds) [09:18:44] *** Quits: tomzawadzki (tomzawadzk@nat/intel/x-bcigpypgviscuzgl) (Ping timeout: 256 seconds) [09:29:20] drv, well crap. I'm gonna have to fix this bdev split base crap before I can add the PT vbdev into the tests... oh well. will let ya know if I get stuck [10:40:14] *** Quits: toka (86bfdc49@gateway/web/freenode/ip.134.191.220.73) (Ping timeout: 260 seconds) [12:17:34] *** Joins: Jones__ (0cda5282@gateway/web/freenode/ip.12.218.82.130) [12:22:27] New to SPDK and going over the documents. Can a filesystem (say ext4) be put over the local NVMe drive? [12:23:47] Jones__: sure. do you have any specific needs? [12:25:00] No. Just curious on how to deploy SPDK. So will a /dev bock device be created for the drive? [12:25:47] not really. it can - but accessing it via kernel block device is costly [12:27:36] Will a userspace block device entry be created? [12:27:45] yes [12:27:52] spdk devices are directly accessible via C API [12:28:26] and spdk as a project delivers a couple of application built upon this api [12:28:39] that is: nvmf, iscsi and vhost servers [12:28:56] And can I use all the standard tools such a fdisk, mkfs.ext4 be used? [12:29:30] you can easily expose SPDK userspace block device to the kernel via nbd [12:30:22] so that the device can be accessed via /dev/nbdX and can be partitioned using standard linux tools [12:31:25] Go it. Will there be performance hit if using nbd? [12:32:37] the performance won't get any better than using linux nvme drivers [12:33:08] Ok. [12:34:44] You mentioned - spdk as a project delivers application such as nvmf and iscsi. What will be a typical deployment model of these two applications? Will these be also through nbd? [12:35:54] they are just userspace applications [12:37:58] if you wanted to encorporate spdk into your existing application that heavily relies on I/O - e.g. a samba server - you would have to modify & recompile it [12:38:06] Do you mean the drives seen on the nvmf and iscsi client will need to be be directly accessed by the userspace application using the SPDK apis? [12:38:44] Understood. [12:39:48] Appreciate your answers on this. It is clear now. [12:40:34] you can read more on spdk.io/doc [12:41:07] http://www.spdk.io/doc/about.html and http://www.spdk.io/doc/userspace.html [13:07:33] Thank you [13:42:15] *** Quits: Jones__ (0cda5282@gateway/web/freenode/ip.12.218.82.130) (Ping timeout: 260 seconds) [14:17:49] hmm, something is going on with the vhost tests within the last few commits [14:17:59] "vhost.c: 727:spdk_vhost_dev_unregister: *ERROR*: Controller naa.Malloc0.1 has still valid connection." during remove_vhost_controller RPC [14:22:51] the VM must be still up for some reason [14:23:15] is it possibly the nullglob patch? [14:23:40] none of the other ones before the issue started seem related [14:29:43] for sure I think we need to revert the migration test [14:30:23] both released builds failed the migration test [14:31:35] https://ci.spdk.io/spdk/builds/release/master/4316/fedora-08/build.log also failed for seemingly the same reason, and that was before the live migration test was enabled [14:32:03] ^ [14:32:16] but immediately after the nullglob patch [14:32:22] so that is my prime suspect currently [14:42:29] vm_list_all() returns just the vm numbers, not the full path [14:42:51] but line 404 isn't prepending the $VM_BASE_DIR now [14:43:23] the basename on line 396 is a nop too [14:43:48] i'll put a patch together [14:47:46] jimharris: I fixed up the whitespace problem on https://review.gerrithub.io/#/c/399786/ and it erased your +2 - otherwise looks good to me [14:49:03] vhost fix: https://review.gerrithub.io/#/c/399951/ [14:49:56] looks good [14:50:16] we'll see if the test pool agrees :-) [14:51:21] hopefully we won't get any Timouts [14:56:49] dude, this is just a tangled hairy mess looking at it the first time :) (bdev unregistration, destruction, hotremove amongst a pile of bdevs) Might have to crack a beer early today! [15:01:56] looking at that bdev ASAN issue? [15:08:20] jimharris: looks like it passed, but also took a minute longer than before (maybe that's just a fluke?) [15:12:04] drv, yeah, we end up in spdk_bdev_part_base_hotremove() with a NULL base_bdev after successfully tearing down the legitimate ones... [15:12:34] hmm, that's definitely not supposed to happen [15:13:33] running it through the pool again [15:14:15] yeah, just in the bdevio test there are 17 legit calls to spdk_bdev_part_base_construct(), 10 to spdk_bdev_part_construct() and they all clean up nicely. Somewhere in there though there's a list not protected or something being freed early or something... will figure it out after some more time [15:14:23] actually I think this is expected - this is the extra time for the migration test [15:14:44] ah, right, I guess it never successfully ran yet [15:22:35] drv, and I suspect that just adding another vbdev into the stack for those tests is making exposing that ASAN issue in other ways... [15:37:20] jimharris: your second run also passed - I say we check it in to fix up the tests, and if pwodkowx has feedback, we can do another patch [15:38:45] pwodkowx already +1'd it [15:39:07] oh, good [15:39:44] carrying a Gerrit-triggered pager or something? :) [15:52:11] *** Joins: Shuhei (caf6fc61@gateway/web/freenode/ip.202.246.252.97) [16:05:53] drv: in https://review.gerrithub.io/#/c/398904/13/lib/nvmf/ctrlr.c - can't we just always use spdk_thread_send_msg and simplify the code? [16:06:13] yes [16:29:59] Hi Jim, Daniel, I would like to hear your thought quickly about how to create JSON config file easily. [16:30:01] https://trello.com/c/pfzHMPGB/24-create-json-config-file-from-json-rpc-dump [16:30:14] Which is better or another option? [16:30:54] Would you spend a minute to it? [16:34:30] I'll do the first option and push to Gerrit. Thanks. [16:44:50] drv, Shuhei: two patches which help reduce test time on fedora-08 (longest system in the pool): https://review.gerrithub.io/#/c/399959/ [16:46:10] OK, please give a time to look into it. [16:49:58] Hi Jim, I think I saw similar idea in other patch. Daniel's input is needed but looks nice to me. [16:52:32] seems OK at first glance [16:52:47] what happens if the NVMe namespace is less than 50 GB already? [16:53:34] hmmmm - yeah, right now this only runs on real NVMe devices [16:53:45] I'm not sure how big the emulated NVMe disks in the VMs are [16:53:49] I could just split it into 4 equal parts [16:54:10] does split even check that the size fits? [16:54:21] it looks like it does [16:54:33] 4 equal parts would probably be OK [17:00:27] *** Quits: lhodev (~Adium@66-90-218-190.dyn.grandenetworks.net) (Read error: Connection reset by peer) [17:00:42] *** Joins: lhodev (~Adium@inet-hqmc01-o.oracle.com) [19:39:42] *** Quits: lhodev (~Adium@inet-hqmc01-o.oracle.com) (Remote host closed the connection) [19:41:19] *** Joins: lhodev (~Adium@66-90-218-190.dyn.grandenetworks.net)