File size: 9,712 Bytes
4aa5fce |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 |
=== _mup__ is now known as _mup_ === zz_mwhudson is now known as mwhudson [22:12] <fwereade> waigani___, heyhey [22:12] <waigani_____> fwereade: hello :) [22:13] <fwereade> waigani___, how's it going? [22:13] <waigani_____> getting there [22:13] <fwereade> waigani___, sorry I haven't really said hi properly since you joined :( [22:13] <waigani_____> a LOT of code to get my head around! [22:13] <fwereade> waigani___, ha, yes indeed [22:13] <waigani_____> no problem, totally understandable [22:13] <waigani_____> ditto, sorry for not saying hi :) [22:14] <waigani_____> so, I'm getting racy hey? [22:14] <fwereade> waigani___, I'm kinda going to bed shortly but I just wanted to chat briefly about the set environ config stuff [22:14] <waigani_____> sure [22:17] <waigani____> fwereade: hangout or irc? [22:17] <fwereade> waigani___, the issue is mainly that I don't really want it to be possible to set an invalid environ config, and I am not convinced that it's possible to guarantee the sanity of a given change without (1) validating the change from the previous version --which is fine, you do it-- and (2) asserting in state that the previous version, from which the change is known to be valid, still applies [22:17] <fwereade> waigani____, I think I can irc it, I will be off to bed in a mo [22:18] <waigani____> ok [22:18] <fwereade> waigani____, does that make sense at first reading or shall I expand a little? ;) [22:19] <waigani____> right, so the real issue is the last one [22:19] <waigani____> how do you ensure the oldconfig is itslef valid [22:19] <fwereade> waigani____, if you can guarantee every change is valid, I think yu can guarantee that the previous version was [22:20] <fwereade> waigani____, because you cannot bootstrap but with a valid config [22:20] <waigani____> can we? [22:20] <waigani____> right [22:20] <waigani____> in which case the current method works then? [22:21] <waigani____> we can rely on an oldconfig [22:21] <waigani____> oohhhh [22:21] <waigani____> lol [22:21] <waigani____> i see [22:21] <waigani____> THAT is why you don't want to allow setconfig without validating [22:22] <fwereade> waigani____, the current method is (at least in theory) crack, because the *actual* change we end up making is pretty random -- see state/state.go:264 [22:23] <waigani____> so this is the race condition you mentioned? [22:24] <fwereade> waigani____, in *practice* it's usually ok, but that's because we don't get many concurrent updates to environ config [22:24] <fwereade> waigani____, yeah, I think so [22:24] <fwereade> waigani____, I have a tendency to froth about a wide range of possible race conditions [22:24] <waigani____> heh [22:25] <fwereade> waigani____, ah, but, yes, that's the one [22:25] <fwereade> waigani____, so what's interesting is that axw is doing some work that involves using an Environ inside state [22:25] <waigani____> what does axw mean when he says The settings changes are actually applied as a delta to what's on disk [22:26] <waigani____> "applied as a delta"? [22:26] <fwereade> waigani____, there's a Settings type [22:26] <fwereade> waigani____, it was written in python for zookeeper and ported straight without much consideration for whether it still makes sense [22:26] <fwereade> waigani____, hint: it doesn't [22:26] <waigani____> lol, okay... [22:27] <fwereade> waigani____, I am struggling to remember the *exact* details [22:27] <fwereade> waigani____, but it is itself fundamentally racy [22:27] <waigani____> don't worry, I'll follow up [22:27] <fwereade> waigani____, it reads existing state, calculates delta, applies delta [22:27] <waigani____> now I know the direction to look in [22:28] <fwereade> waigani____, pretty sure it doesn't assert the base state still holds [22:28] <waigani____> now state == mongodb? [22:28] <fwereade> waigani____, yeah, take a look at that for a bit of context [22:28] <fwereade> waigani____, yep [22:28] <waigani____> i.e. config stored in mdb [22:29] <fwereade> waigani____, but have a word with axw, because *he* is doing some work that involves using an Environ inside state to check sanity of AddMachine ops by checking with the provider [22:29] <fwereade> waigani____, and *you* are doing something that involves checking sanity of applying changes to Environ (configs) [22:30] <waigani____> will do. [22:30] <waigani____> would you like to see the ApplyAndValidate method I've refactored out, added back into SetEnvironConfig? [22:31] <waigani____> Such that you cannot set an environ config without validating it? [22:31] <fwereade> waigani____, I think that is at least part of it, but be suspicious of what actually happens when you try to use that settings type [22:32] <waigani____> fwereade: be suspicious of SetEnvironConfig? [22:32] <fwereade> waigani____, it may be that you're best off using entirely different code that just happens to use the same data format [22:32] <fwereade> waigani____, more of state.Settings [22:33] <fwereade> waigani____, ...or, heh, maybe converting it to use a new data format [22:33] <fwereade> waigani____, I'm also pretty sure that the fields are dumped straight into the mongo document [22:33] <waigani____> fwereade: just saw your comment: TODO(fwereade) state.Settings is itself really problematic in just about every use case [22:33] <fwereade> waigani____, and by sheer conincidence magic fields like txn-revno happen not to collide with actual env settings [22:34] <fwereade> waigani____, I think what I'm saying is that you have stumbled into a rabbit hole [22:34] <fwereade> waigani____, do not feel obliged to fix everything [22:34] <waigani____> I was just about to say the same! [22:35] <fwereade> waigani____, but be careful to understand what you're fixing and what you're not, and please document it clearly [22:35] <fwereade> waigani____, take heart: you're unlikely to make that specific code *worse* [22:35] <waigani____> obviously my understanding is cursory, so I'll follow up with axw and look into state.Settings [22:35] <fwereade> waigani____, and I'll be happy if it just gets a little bit better [22:36] <fwereade> waigani____, awesome, thanks [22:36] <waigani____> are there any other hits for me? [22:36] <waigani____> bits of code I should look at? [22:36] <fwereade> waigani____, for this particular stuff it's quite localised [22:37] <fwereade> waigani____, there are a couple of other Settings clients [22:37] <fwereade> waigani____, one day we'll get rid of them all [22:37] <waigani____> okay [22:37] <fwereade> waigani____, but it's the env config that really upsets me [22:38] <fwereade> waigani____, because it's the most potentially horrible one ;) [22:38] <fwereade> waigani____, anyway I should get some sleep [22:38] <waigani____> just before you go .... :D [22:38] <fwereade> waigani____, I might be a little late tomorrow, but dimitern is staying here, so if you see him and want to talk to me tell him to come shout at me :) [22:39] <fwereade> waigani____, I'm still here :) [22:39] <waigani____> could you give me one example of the race conditon? [22:39] <waigani____> two clients trying to set the config at the same time? Is that right? [22:39] <fwereade> waigani____, yeah [22:40] <waigani____> clients = machine agent, juju cli, ... [22:40] <fwereade> waigani____, ah! [22:40] <fwereade> waigani____, there *is* a replaceSettingsOp func [22:40] <fwereade> waigani____, that I think is not racy [22:40] <waigani____> oh [22:40] <fwereade> waigani____, it may need to be retried [22:40] <fwereade> waigani____, but if you use it the txn will abort [22:41] <waigani____> okay, lots to learn ... I'll let you sleep [22:41] <fwereade> waigani____, so, *if* you can construct a known-valid config, you can use replaceSettingsOp to be sure it actually lands in state [22:41] <waigani____> thanks for taking the time to walk me through the rabbit whole, a little ;) [22:41] <fwereade> waigani____, but to construct a known-valid one inside state, you'll need to talk to axw [22:41] <waigani____> *hole [22:41] <fwereade> waigani____, because [22:41] <fwereade> (you're sitting down, right) [22:42] <waigani____> hehe yep [22:42] <fwereade> waigani____, just because it's a valid config.Config [22:42] <fwereade> waigani____, it is *not* necessarily a valid config for any given provider [22:42] <waigani____> ugh [22:42] <fwereade> waigani____, let alone a valid change for that specific environ [22:42] <waigani____> what? [22:43] <waigani____> I get that it might not be valid for other providers [22:43] <fwereade> waigani____, simplest case: some fields are immutable, or should be [22:43] <fwereade> waigani____, change an env's name from foo to bar, bad things will happenb [22:43] <fwereade> waigani____, constructing an env, and setting the new config, *should* catch all those cases [22:43] <waigani____> so that should be caught as invalid by the environ provider validator? [22:44] <waigani____> right [22:44] <fwereade> waigani____, *but* the environs package uses the state package, so you can't directly reference an Environ as such from code inside state [22:44] <fwereade> waigani____, axw is dealing with *exactly* the same problem, creating environs inside state to validate machine addition [22:45] <fwereade> waigani____, so it would be good if your solutions were roughly aligned :) [22:45] <fwereade> waigani____, was that a bit more helpful? [22:46] <waigani____> yes, you've set me in the right direction [22:46] <fwereade> (ok, it's not *exactly* the same problem, but you can surely talk productively) [22:46] <waigani____> I'll go annoy axw now ;) [22:46] <fwereade> waigani____, cool, cheers [22:47] <waigani____> thanks again, sleep well [22:47] * fwereade disappears |