[01:00:14] *** Quits: johnmeneghini (~johnmeneg@pool-96-252-112-122.bstnma.fios.verizon.net) (Quit: Leaving.) [01:03:10] *** Joins: gila (~gila@5ED4FE92.cm-7-5d.dynamic.ziggo.nl) [03:21:33] *** Quits: gila (~gila@5ED4FE92.cm-7-5d.dynamic.ziggo.nl) (Quit: My Mac Pro has gone to sleep. ZZZzzz…) [03:30:48] *** Joins: gila (~gila@5ED4FE92.cm-7-5d.dynamic.ziggo.nl) [05:31:31] dcostantino, I haven't heard of anyone jumping on it yet... [05:32:59] it's come up before, just haven't heard of anyone starting the work. I added a trello card for it at https://trello.com/b/P5xBO7UR/things-to-do [05:33:08] is it something you're interested in doing? [06:33:53] *** Joins: lhodev (~Adium@inet-hqmc08-o.oracle.com) [07:08:00] *** Quits: gila (~gila@5ED4FE92.cm-7-5d.dynamic.ziggo.nl) (Quit: My Mac Pro has gone to sleep. ZZZzzz…) [07:09:59] *** Joins: gila (~gila@5ED4FE92.cm-7-5d.dynamic.ziggo.nl) [08:40:37] *** Joins: pwodkowx (pwodkowx@nat/intel/x-mhhmhcsiswxzpfsz) [08:40:59] Hi, can someone look at https://review.gerrithub.io/#/c/374462/ ? [08:41:28] recent patch broke setup.sh c8328f557c23 ("setup.sh: improve class code and dev/vendor ID checking") [08:41:48] looking... [08:43:14] I try to fix it but seem there is ancient version of lspci tool on Centos 6 and I don't know awk enough to fix the issue. [08:45:31] ok - maybe we should just revert this setup.sh patch for now [08:52:54] pwodkowx: so on your system if device id starts with "0e", that PCI BDF will always get printed? [08:53:17] yes, and not only on my [08:53:47] pawelk and lukaszg also have this issue. [08:54:53] dariusz suggested that this might be some kind of implicit conversion to number or something (scientific notation?) [08:58:40] $ echo 0e19 | awk -v id="0e02" '{ if (id == $1) print $1 " equals " id }' [08:58:51] 0e19 equals 0e02 [08:58:53] but [08:59:15] echo 0eXX | awk -v id="0e02" '{ if (id == $1) print $1 " equals " id }' [08:59:19] is working [09:03:06] in the current version of the script, can you try: [09:03:14] lspci -mm -n -D | tr -d '"' | awk -v cc="$1" -F " " '{if (cc ~ $2) print $1}' [09:03:21] basically replace "==" with "~" [09:03:35] and: [09:03:46] lspci -mm -n -D | tr -d '"' | awk -v ven="$1" -v dev="$2" -F " " '{if (ven ~ $3 && dev ~ $4) print $1}' [09:04:02] *** Parts: lhodev (~Adium@inet-hqmc08-o.oracle.com) () [09:09:41] https://review.gerrithub.io/#/c/374507/ [09:11:45] *** Quits: pwodkowx (pwodkowx@nat/intel/x-mhhmhcsiswxzpfsz) (Ping timeout: 248 seconds) [09:12:28] Hello all, quick question: Do you know where in the spdk the DPDK is being de-initialized? I think Ive got my implementation of blobstore working and am trying to run tests using google test. The first test passes (simple init and de-init) however, the second test automatically fails due to DPDK not initializing correctly. My guess from playing around for a while now is that the DPDK is still initialized from the [09:12:28] first test and is causing the second test to fail. [09:12:28] basically the second tests fails no matter what the first test is executing because it tries to re-initialize the DPDK. My guess is that im not deinitializing properly [09:13:33] hi nKumar - are your first test and second test all running within the same test app? [09:14:07] DPDK gets initialized during the call to spdk_env_init() [09:14:46] this should get called just once, at the very beginning of your test application (not before each test within that application) [09:20:18] jimharris: your patch series starting here needs to be rebased: https://review.gerrithub.io/#/c/374507/ [09:20:34] (needs "test/bdev: remove partition table after GPT test") [09:20:50] I believe google test actually splits each test up into separate main functions. So in the second test when the blobStore class I have calls spdk_env_init() it fails. [09:21:03] drv: thanks - done [09:21:44] nKumar: does google test allow you to just call spdk_env_init() once upfront, instead of once for each test? [09:22:21] dcostantino: I started playing around with implementing NVMe 1.3 directives (i.e. streams), but I didn't get very far [09:22:37] if someone was willing to jump in and do it, it would be very much appreciated [09:24:10] @jimharris, I dont believe so. I should be able to control this via a parameter passed to my init function though. Let me give that a shot. [09:51:38] bwalker, so I need to update almost every patch in my chain and looking at directions you gave me before I believe it’s all clear except one step. Briefly (1) checkout out hash (2) make tmp branch, make changes, —amend. At this point do I push or do I go straight to checking out the last patch in the series and doing rebase -i? And if the later, what actions do I take on the rebase for each patch listed? Thanks!! [09:51:54] (no hurry, not going to mess with it til maybe tomorrow anyway) [09:52:18] or (3) use stgit [09:52:21] :) [09:52:25] I think for big chains like this, you want to either 1) ask Jim how to use stgit or 2) use rebase -i [09:52:31] jimharris, yeah, that's next once I get this process down! [09:52:40] oh no :) [09:52:42] do you'd check out the top of your chain [09:52:46] do git rebase -i master [09:53:00] edit the proposed plan to change each revision from "pick" to "edit" [09:53:10] then it will zoom you to the first commit - make your changes, git commit --amend [09:53:13] then git rebase --continue [09:53:18] it will jump to the next commit to be modified [09:53:20] do the same thing [09:53:23] until it is over [09:53:35] so after I ammend on the tmp branch, do I push or go straight to checking out the tip? [09:53:46] and doing the -i process like you just outlined [09:54:12] the way I'm telling you to tackle this doesn't actually involve making an explicit tmp branch [09:54:27] you do these commands: [09:54:32] git checkout [09:54:35] git rebase -i master [09:54:44] oh, just start with checking out the earliest patch I need to edit then? [09:54:46] edit that file in a text editor, change "pick" to "edit" for the commits that should be modified [09:56:13] because when you say checkout "your branch" I'm not sure whether that means the first patch (oldest) or last (newest)? [09:56:32] your changes are all on one branch, aren't they? [09:56:35] with some name [09:56:45] you don't have a branch pointer for each commit I assume? [09:58:22] yeah, they're all on one and right now I envision it pointing to the last in the series since that's the last thing I pushed. That's how I think about it anyway [09:58:39] yeah - so your branch pointer is pointing at the last commit on the branch [09:58:45] so git checkout [09:58:51] moves you to that last commit on your branch [09:58:56] as opposed to checking out the hash of the first patch in the series that I get from GH which I've done in the past which may be the root cause for my breaking the chain all the time [09:59:17] or... I can pay you a hundred bucks to do it all for me ;) [09:59:32] I like that hourly rate [09:59:33] "st bwalker" [10:00:27] OK, so I think I got it, one more question. After doing al the edits/amends is there just one push at the very end then? [10:00:51] yep - if you push the last commit on your branch, it's smart enough to push all of the ones before it as necessary [10:01:47] *** Joins: lhodev (~Adium@inet-hqmc08-o.oracle.com) [10:02:09] *** Joins: johnmeneghini (~johnmeneg@pool-96-252-112-122.bstnma.fios.verizon.net) [10:03:29] rock-n-roll. thanks [10:03:32] *** Quits: johnmeneghini (~johnmeneg@pool-96-252-112-122.bstnma.fios.verizon.net) (Client Quit) [10:12:21] While also relatively new to SPDK, I'm a complete noob to GerritHub. I've created an GitHub ID, uploaded ssh key, etc. and brought up Daniel's patch, https://review.gerrithub.io/#/c/374371/. I've navigated about that page including walking through the individual files that were changed. I see at the bottom of the page under "History" for Jim Harris: Patch Set 1: Code-Review+2. When I click on that, it merely expands a line. There's [10:12:21] no additional info. Does that mean Jim had no additional comments? That is, does this imply Jim clicked on '+2 Looks good to me, approved" without any additional "Cover Message" nor inline code comments? [10:18:17] correct [10:18:56] Very well. Thx Ben. [10:20:31] you only need to sign in and upload an ssh key if you plan to submit patches (and I hope you do) [10:21:00] I guess you have to do the github account import to vote on patches too [10:46:54] *** Joins: johnmeneghini (~johnmeneg@pool-96-252-112-122.bstnma.fios.verizon.net) [10:46:55] *** Quits: johnmeneghini (~johnmeneg@pool-96-252-112-122.bstnma.fios.verizon.net) (Client Quit) [11:03:48] Hi team, If someone is looking for SPDK to expose a generic block device that they can put a file system on. Is this something they can do with Logical Volumes [11:06:27] put a Linux kernel filesystem on? [11:06:52] *** Quits: dcostantino (~dcostanti@c-24-23-244-114.hsd1.ca.comcast.net) (Quit: leaving) [11:07:46] They are using XFS [11:09:24] lhodev, and also, FYI, they way the +/- system works... only a maintainer and +2 something but everyone is encouraged to review and do a +1/-1 to provide their input. After 2 +2 are achieved the patch will be merged to master... the maintainers take the +/-1 into account but they may not block a merge depending on how valid the comments are [11:11:06] *** Joins: johnmeneghini (~johnmeneg@pool-96-252-112-122.bstnma.fios.verizon.net) [11:11:07] *** Quits: johnmeneghini (~johnmeneg@pool-96-252-112-122.bstnma.fios.verizon.net) (Client Quit) [11:12:16] jkkariu: can you explain more? this works today with iSCSI, NVMe-oF or vhost - for example, with NVMe-oF, the kernel on the initiator connects to the SPDK NVMe-oF target, and will then present a kernel block device that can be formatted with any filesystem [11:12:54] similar for vhost - the block device(s) are exposed through the guest VM kernel, where any filesystem can be formatted [11:15:12] if you want to put XFS on a local SPDK block device, the only "option" is nbd - we have some support for this currently but performance is really bad since it requires a full data copy between the kernel and the SPDK process managing the userspace block device [11:15:53] currently we use this more for control path type activities - like using standard Linux partitioning utilities to enable SPDK GPT testing [11:16:16] Jim, thanks for the information. They are using MySQL DB. The storage is local. No virtualization [11:19:16] MySQL with local storage with InnoDB writing to XFS is not something we have much of a solution for [11:19:40] however, if they are using MySQL with MyRocks on XFS, then they can use our RocksDB backend to bypass XFS and the kernel [11:20:02] that's still sort of proof of concept, but we definitely have MySQL running with our RocksDB [11:22:06] sethhowe: your 5 second delay - do we have any VMs using soft RoCE yet? [11:22:45] if this is in preparation for adding VMs with soft RoCE, then I'm OK with a 5 second delay (instead of 2) if we think that's needed to avoid any intermittent failures [11:23:10] I thought this was to get rid of some existing intermittent failures with real RDMA NICs [11:24:46] has anyone else noticed that the error logs for CI failures don't seem to be getting populated lately (even after like 24 hrs)? Seems like it's all patch failures, not just mine [11:30:59] peluse: do you mean on ci.spdk.io? [11:31:25] the rsync that copies stuff to the public ci.spdk.io server gets way behind - we are working on cleaning up old review builds so it should get faster [11:39:34] jimharris: yes, we do. But it looks like the 2 second delay fixes it. [11:39:35] drv, yeah. Never seen it take so long so figured something was up. thanks [11:40:47] our bandwidth to copy to that web server is pretty limited [11:42:17] * peluse envisions little mice carrying messages from one machine to another through a minefield of cheese [11:42:37] I would complain, but OTC is hosting that for us without charging us anything [11:42:51] hard to beat free! [11:50:46] jimharris fedora-06 is the machine currently using soft-roce. We ran the tests about 5 times with a two second delay without hitting that intermittent failure. [13:17:33] jimharris, bwalker: I fixed up a merge error in the vhost common features patch - please take another look: https://review.gerrithub.io/#/c/373577/ [13:20:33] good catch - done [15:20:15] the blobstore CLI is available for review with a decent initial set of function described in the commit msg. It's a lot for one review but mostly cookie cutter stuff similar to hello_blob so sorry about the size but take your time... https://review.gerrithub.io/#/c/373383/ [15:26:00] *** Quits: patrickmacarthur (~smuxi@2606:4100:3880:1240:2977:4084:1d52:877) (Remote host closed the connection) [16:21:26] drv: working on this virtio-scsi initiator bdev module a bit - I need virtio_scsi.h (similar to vhost) - do you see any problems if I move the internal copy we have in lib/vhost/linux_virtio_scsi.h and move it to include/linux/virtio_scsi.h? [16:27:10] jimharris: sounds OK to me [16:55:50] *** Quits: lhodev (~Adium@inet-hqmc08-o.oracle.com) (Quit: Leaving.) [21:42:08] *** Quits: pzedlews (~pzedlews@192.55.54.44) (K-Lined) [21:51:21] *** Joins: bwalker (~bwalker@192.55.54.44) [21:51:21] *** Server sets mode: +cnt [21:51:21] *** Server sets mode: +cnt [21:54:23] *** Joins: jimharris (~jimharris@192.55.54.44) [21:54:54] *** Joins: ziyeyang (~ziyeyang@192.55.54.44) [21:55:25] *** Joins: qdai2 (~qdai2@192.55.54.44) [21:56:25] *** Joins: vermavis (~vermavis@192.55.54.44) [21:59:26] *** Joins: pbshah1 (~pbshah1@192.55.54.44) [21:59:57] *** Joins: gangcao (~gangcao@192.55.54.44) [22:01:02] *** Joins: kjakimia (~kjakimia@192.55.54.44) [22:03:34] *** Joins: pzedlews (~pzedlews@192.55.54.44) [22:04:04] *** Joins: mszwed (~mszwed@192.55.54.44) [22:04:48] *** Joins: jkkariu (~jkkariu@192.55.54.44) [22:07:31] *** Joins: peluse (~peluse@192.55.54.44) [22:08:32] *** Joins: cunyinch (~cunyinch@192.55.54.44) [22:10:03] *** Joins: ppelplin (~ppelplin@192.55.54.44) [22:19:06] *** ChanServ sets mode: +o jimharris