[00:42:20] *** Joins: tomzawadzki (uid327004@gateway/web/irccloud.com/x-qykgfijquwnjuakl) [02:34:56] @peluse - as for the nvme_phy_autotest there was something wrong with nvme-cli test dependencies. /home/sys_sgsw/nvme-cli/spdk was a directory instead of a symlink do spdk sources, looks fixed at the moment [02:35:24] This was after I updated the systems to FC28 so probably my overlook [02:36:34] Failing nightly vhost tests indeed look similar on Jenkins and CTP. Seems like live migration test case is failing. Asked pniedzwx to investigate. [03:46:26] p *(int*)0x7ffff16f2800 [03:46:53] oops, wrong window [03:47:14] unless you can tell me what's under 0x7ffff16f2800 [03:57:01] *** Joins: tkulasek (~tkulasek@134.134.139.83) [04:21:59] *** Joins: felipef (~felipef@cpc92310-cmbg19-2-0-cust421.5-4.cable.virginm.net) [04:31:17] *** Quits: felipef (~felipef@cpc92310-cmbg19-2-0-cust421.5-4.cable.virginm.net) (Ping timeout: 244 seconds) [04:36:52] *** Joins: felipef (~felipef@cpc92310-cmbg19-2-0-cust421.5-4.cable.virginm.net) [04:52:47] *** Quits: felipef (~felipef@cpc92310-cmbg19-2-0-cust421.5-4.cable.virginm.net) (Ping timeout: 240 seconds) [05:21:30] *** Joins: felipef (~felipef@cpc92310-cmbg19-2-0-cust421.5-4.cable.virginm.net) [06:53:05] klateck, thanks!!! [06:53:27] 0x7ffff16f27fff is just under :) [06:55:03] that's the answer i needed [06:55:17] * peluse is here to help [07:19:07] *** Quits: tkulasek (~tkulasek@134.134.139.83) (Ping timeout: 240 seconds) [08:13:15] *** Quits: felipef (~felipef@cpc92310-cmbg19-2-0-cust421.5-4.cable.virginm.net) (Remote host closed the connection) [08:21:46] jimharris: about notification functionality me ant piotr are working on. Is this needs to be a separate library? [08:23:37] w have problems about dependency during link time. notification code itself neds to modofy many makefiles and RPC part is adding RPC and JSON library dependencies to almost every component in SPDK [08:26:49] can we move it to lib/event? [08:29:37] if we put it in lib/event, then all modules which send notifications will be dependent on the event framework [08:30:11] for sure [08:30:28] and we've wanted to avoid that - for example, bdev layer is not dependent on the event framework today, and the bdev fio plugin is especially designed to show how to use it outside of the event framework [08:31:24] alternative is that we me make every module depended from thee new libraries: lib/notify, lib/rpc and lib/json [08:31:26] can we put just the RPCs in lib/event/notify? [08:32:04] i was thinking that the RPCs for getting notifications would be separate from the core notify logic [08:32:17] meaning an RPC is just one type of possible client for receiving notifications [08:36:40] we can, but this will also force these 3 libs dependency for every app that use event framework. [08:38:41] are we ok with that? [08:38:54] are there any apps today that use the event framework but don't already link in lib/rpc and lib/json? [08:38:55] Maybe we could make default implementation of notifications to redirect events to /dev/null. [08:39:28] *** Joins: travis-ci (~travis-ci@ec2-54-205-36-179.compute-1.amazonaws.com) [08:39:29] (spdk/master) app/trace: fix app_name dereference on init (Tomasz Zawadzki) [08:39:29] Diff URL: https://github.com/spdk/spdk/compare/3126157a2192...c3c69229414b [08:39:29] *** Parts: travis-ci (~travis-ci@ec2-54-205-36-179.compute-1.amazonaws.com) () [08:39:39] i was thinking that if some module sends a notification, but there are no active clients listening for notifications, that the notification just gets dropped on the floor [08:39:51] which is effectively /dev/null, but it's implicit [08:40:50] yes, and this is how Piotr started implenting this [08:42:26] are there any apps today: fuse is one [08:45:21] weird bug. I added spdk_strerror() to env_dpdk and apparently it broke the hello_world nvme app [08:45:28] undefined spdk_strerror reference [08:46:02] wrapping the -lspdk_util lib (which spdk_strerror is from) with --whole-archive fixes that [08:49:39] i was calling cc on hello_world manually. It couldn't find spdk_strerror with just -lspdk_util [08:50:24] instead of adding --whole-archive, can you edit mk/nvme.libtest.mk instead? [08:50:39] and put ENV_LINKER_ARGS first in LIBS [08:50:41] instead of last [08:51:11] we could also make that each app idepend on all SPDK libs this will fix both issues :D [08:54:00] jimharris: ah right, forgot about the ordering [08:54:09] that'll probably fix it [08:56:21] pwodkowx: nbd.c: 928:spdk_nbd_start: *ERROR*: ioctl(NBD_SET_SOCK) failed: Device or resource busy [08:56:27] that's a fresh one from CTP [08:56:46] we've seen those before [08:57:33] darsto: i saw your comment about removing --whole-archive when linking the unit tests - I can certainly put it back in (removing it in a separate patch is probably better), but i really don't think it is needed for unit tests [08:58:26] won't it break our lto build? [08:59:24] i'm not following [08:59:33] *** Joins: felipef (~felipef@cpc92310-cmbg19-2-0-cust421.5-4.cable.virginm.net) [08:59:52] --whole-archive is only there to make sure that constructor functions don't get optimized out [09:00:29] let me check [09:00:36] i'm not aware of any unit tests that are depending on constructor functionality [09:01:19] tests like subsystem_ut.c explicitly call spdk_add_subsystem - it doesn't rely on the constructors, since it needs to set up and tear down between each test [09:01:19] *** Quits: felipef (~felipef@cpc92310-cmbg19-2-0-cust421.5-4.cable.virginm.net) (Read error: Connection reset by peer) [09:02:44] LINK test/unit/include/spdk/histogram_data.h/histogram_ut [09:02:44] :(.text+0xa79): undefined reference to `__wrap_calloc' [09:03:33] that's from enabling lto? [09:03:36] that's not a constructor, but still requires --whole-archive apparently [09:03:38] yep [09:05:23] ok - then i guess we just leave it in for now [09:07:42] jimharris: you left out one \ at the end [09:08:02] at the end of SPDK_STATIC_LIB_LINKER_ARGS [09:08:39] you mean i added an extra one? [09:09:17] it was there, and it was correct in the previous context, but now it's not [09:09:34] just a nitpick while you're at it [09:10:01] i just pushed a new version [09:14:46] still breaks lto though [09:14:52] ordering issue? [09:19:31] you mean even with whole-archive, it still doesn't work? [09:20:02] yes, this time it's LINK test/env/pci/pci_ut [09:20:03] :(.text+0x9519): undefined reference to `spdk_mempool_create' [09:23:05] can you try adding __attribute__((used)) to DECLARE_WRAPPER? [09:23:09] (HEAD^ compiles just fine) [09:23:16] hmmm [09:25:21] error: ‘used’ attribute ignored [-Werror=attributes] [09:25:31] it can't be used with declarations, i guess [09:31:53] trying it on my system now [09:32:02] (lto, not the used attribute) [09:34:15] pci_ut includes "env_dpdk/pci.c", but still links with env_dpdk [09:34:46] could that be an issue? [09:35:29] jimharris, FYI I have the comp vbdev coded to test a single IO direct to DPDK (local, not pushed). If can test/debug before tomorrow morn, I'll ping ya about integrating reduce. Depending on how much time I have it might have to wait til after turkey though.... will let ya know! [09:35:33] on my system, with my patch, if i leave in the --whole-archive, I get a whole slew of other multiple definition errors [09:38:12] darsto: ok - I think I have a fix for this [09:38:19] i'll push something shortly [09:39:03] peluse: cool [09:40:06] jimharris: nice, i don't have a clue [10:25:06] bwalker, fyi I replied to your comment on https://review.gerrithub.io/c/spdk/spdk/+/429523 - no hurry at all just didn't want it to get lost in the patch history of rebases, etc.... [10:28:57] hmm, are we over engineering this? [10:32:53] bwalker, the configure thing or different conversation? [10:33:01] the two configure flags [10:33:13] possibly... [10:33:48] libreduce is only useful for compression - I don't see how it would work for de-dupe [10:34:43] I'm not sure if libreduce was going to directly depend on the compressdev framework or if it was going to have callbacks for that, but it has to be provided with some compression mechanism [10:35:37] callbacks [10:35:53] it will interface with the vbdev module [10:36:15] I can collapse them into one if you prefer that, jimharris was leaning that direction as well. [10:36:41] but I bought him some dry erase markers to bribe him into using two ;) [10:37:06] is the idea behind two that you think someone may want to compile libreduce but not use the vbdev/dpdk? [10:37:19] yes [10:38:06] at some point in the future maybe possibly potentially. Also just thinking of it as a "standalone" library with its own interfaces, not directly dependent on the vbdev module or DPDK. Someone could write another vbdev module and interface it as well [10:38:25] "on the compression vbdev module" I mean [10:38:33] but if libreduce doesn't have any external dependencies - it just takes in callbacks [10:38:38] then it doesn't need a flag at all, right? [10:38:44] may as well just always build it [10:39:04] and have one flag around the vbdev build, because that has actual dependencies [10:39:12] hmmmm, yeah [10:39:32] jimharris, why did you have it as a config option in your ./configure patch? [10:39:43] * peluse will be away for just a bit.... [10:40:36] libreduce has a direct dependency on PMDK [10:40:45] that was the only reason I had a configure flag for it [10:40:55] could we use the pmdk flag? [10:41:00] I guess that's confusing too though [10:41:21] I'm mostly worried about the user experience when running configure - they aren't going to understand all of the dependencies [10:44:13] *** Quits: tomzawadzki (uid327004@gateway/web/irccloud.com/x-qykgfijquwnjuakl) (Quit: Connection closed for inactivity) [10:52:22] I think for now it will be the best user experience to just have a single flag [10:52:31] that turns on all of the libreduce/compression stuff [10:52:36] if you specify that, it makes sure it can build it all [11:02:14] *** Joins: travis-ci (~travis-ci@ec2-54-160-131-4.compute-1.amazonaws.com) [11:02:15] (spdk/master) config_converter: Fix conversion error to etc/spdk/vhost.conf.in (Shuhei Matsumoto) [11:02:16] Diff URL: https://github.com/spdk/spdk/compare/c3c69229414b...ac7821dc822c [11:02:16] *** Parts: travis-ci (~travis-ci@ec2-54-160-131-4.compute-1.amazonaws.com) () [11:04:09] *** Joins: travis-ci (~travis-ci@ec2-54-80-248-52.compute-1.amazonaws.com) [11:04:10] (spdk/master) ioat/perf: handle spdk_mempool_get() failure (Tomasz Zawadzki) [11:04:11] Diff URL: https://github.com/spdk/spdk/compare/ac7821dc822c...33c517c57a42 [11:04:11] *** Parts: travis-ci (~travis-ci@ec2-54-80-248-52.compute-1.amazonaws.com) () [11:11:56] bwalker, I believe the only one I expose in 'help [11:12:03] ' is the compress optionb [11:12:24] reduce is also exposed in configure [11:13:41] that's what I mean, in the configure script the usage screen doesn't mention reduce. It's a "behind the scenes" option that we turn on when they --with-crompress [11:13:46] users only see one option [11:14:11] I could remove all the reduce flags an jsut set the PMDK flag when the choose compress.... [11:14:12] in your patch you hide reduce? I'm just looking on master and the reduce option is visible [11:14:20] in my patch, yes [11:14:33] master has Jim's initial patch that dosn't include compress [11:14:52] but can't you just piggy-back on the existing reduce flag instead of adding a new one? [11:15:05] that was Jim's suggestion, sure I can do that [11:15:29] they're both in there so quick vote CONFIG_REDUCE or CONFIG_COMPRESS? [11:15:46] reduce is already merged to master, so I vote reduce [11:16:06] OK, will do. tomorrow morn though, going back on vacation now :) Thanks!! [11:16:13] enjoy! [11:16:26] 11 am isn't too early for beer on vacation FYI [11:16:32] so get to it [11:17:06] oh hell yeah!!! :) [11:21:44] *** Joins: travis-ci (~travis-ci@ec2-54-198-115-13.compute-1.amazonaws.com) [11:21:45] (spdk/master) test/iscsi_tgt: Port CHAP to JSON RPC in Calsoft test (Shuhei Matsumoto) [11:21:46] Diff URL: https://github.com/spdk/spdk/compare/33c517c57a42...cd209dbbafce [11:21:46] *** Parts: travis-ci (~travis-ci@ec2-54-198-115-13.compute-1.amazonaws.com) () [12:17:27] bwalker: peluse: I think oatmeal stouts are perfect for breakfast every day. [13:34:10] *** Joins: travis-ci (~travis-ci@ec2-54-211-58-57.compute-1.amazonaws.com) [13:34:11] (spdk/master) jsonrpc client: move around functions so no forwad declaration is needed (Pawel Wodkowski) [13:34:11] Diff URL: https://github.com/spdk/spdk/compare/cd209dbbafce...4734811dc66c [13:34:11] *** Parts: travis-ci (~travis-ci@ec2-54-211-58-57.compute-1.amazonaws.com) () [13:34:40] *** Joins: travis-ci (~travis-ci@ec2-54-221-166-78.compute-1.amazonaws.com) [13:34:41] (spdk/master) vtophys: remove internal device refcount (Darek Stojaczyk) [13:34:41] Diff URL: https://github.com/spdk/spdk/compare/4734811dc66c...04babbea355b [13:34:41] *** Parts: travis-ci (~travis-ci@ec2-54-221-166-78.compute-1.amazonaws.com) () [13:39:50] *** Joins: travis-ci (~travis-ci@ec2-54-204-175-108.compute-1.amazonaws.com) [13:39:51] (spdk/master) nvme: Add the NVMe over fabrics TCP/IP transport support (Ziye Yang) [13:39:51] Diff URL: https://github.com/spdk/spdk/compare/04babbea355b...e956be96eb63 [13:39:51] *** Parts: travis-ci (~travis-ci@ec2-54-204-175-108.compute-1.amazonaws.com) () [15:39:52] *** Quits: klateck (klateck@nat/intel/x-brmsvxqrhptjovne) (Ping timeout: 268 seconds) [15:39:52] *** Quits: stefanha (~stefanha@yuzuki.vmsplice.net) (Ping timeout: 268 seconds) [15:39:52] *** Quits: pwodkowx (~pwodkowx@134.134.139.72) (Ping timeout: 268 seconds) [15:39:52] *** Quits: pniedzwx (~pniedzwx@134.134.139.72) (Ping timeout: 268 seconds) [15:40:20] *** Quits: mszwed (~mszwed@134.134.139.72) (Ping timeout: 268 seconds)