Richard Fairhurst a scris: > D Tucny wrote: >> Eddy Petrișor <eddy.petri...@gmail.com> wrote: >>> What's even more weirder is the fact that if I try to edit with >>> Potlatch, the way is displayed correctly, as if stale data was read of >>> as if Mapnik's data was read. >>> [...] >> There was a post 12 hours ago to the talk list about the exact same >> thing... >> >> Richard the author of potlatch explained the issue in his reply to that >> post 11 hours ago as being a partial written save due to server load >> or network problems or some other weird issue, the result being that >> ways can contain nodes that are deleted...
I am not subscribed to the talk list. Is not exactly clear which lists should be of interest, depending on the usage of the user goals and which are a *must* if one intents to do some type of work, like development. > Exactly (and thanks for posting before I got a chance to :) ). Hello, Thanks for the explanations. > When Potlatch writes a way it first of all updates the nodes: > > http://trac.openstreetmap.org/browser/sites/rails_port/app/controllers/amf_controller.rb?rev=11420#L356 > > then it deletes any nodes that were used by the old version, but not the new > one: > > http://trac.openstreetmap.org/browser/sites/rails_port/app/controllers/amf_controller.rb?rev=11420#L400 > > and finally, it writes the new version of the way: > > http://trac.openstreetmap.org/browser/sites/rails_port/app/controllers/amf_controller.rb?rev=11420#L411 > > What appears to be happening in these cases is that the server is timing out > while deleting the surplus ("unique") nodes. Therefore there is a > discrepancy between the state of the nodes, and of the way. > > I have just committed a change to swap the last two stages around. Therefore > if the server breaks when deleting the unused nodes, the way will already > have been written. This of course is not really fixing the problem, it's > moving it elsewhere, but hey. As I understand it, in the worse case scenario, this could lead to possible dead nodes appearing, right? (Of course, this wouldn't be as bad as having corrupt ways, so it's an improvement.) > On a side issue it is offensive and demoralising when people jump to > conclusions in this way, especially when they clearly haven't bothered to > search the mailing list archives first: this has been discussed several Sorry for not searching the archives, it is indeed my fault, and I am sorry if you found my initial mail offensive. I spite of that, I always advice people not to take criticism of the projects the work on as a personal offensive, but only pick the objective information since, in most cases, people act on either panic (OMG, this is breaking data, I'd better tell people since it might break anything is touched), frustration (damn, this made me waste 3 hours of work) or some other badly charged emotional state and seldom out of pure pleasure. I acted on panic, so please accept my apologies if I managed to negatively charge the discussion from the get go. Looking on the bright side, I still prefer Potlatch to JOSM just because is accessible anywhere and it doesn't make my system crawl. I can start reasonably fast working on stuff I want to correct (without a GPX trace) without trying to spot the area close and small enough for me to get the exact rectangle that interests me (like JOSM does without a planet file). I like the "parallel" way feature and I like that the yahoo! aerials work (as opposed to JOSM with firefox3). OTOH, back to the discussion, I am not subscribed to talk and I didn't knew it was a precondition to joining dev. If this is true, maybe the description of dev should say "OpenStreetMap developer discusssion. Is recommended you also join "talk", if you're subscribing to this list". Please note that I am not trying to act smart, but I am really trying to help and suggest improvements. > times in the past, and as D said, most recently 12 hours ago! It is not a > "really BAD bug" in Potlatch, it is poor server performance allied to the > lack of transactions. It has been explained many times in the past, not just Which leads to what looks like a path for data breakage, as some people on the RO-OSM list feared. Taking into account that the history of the way was truncated and that it was missing nodes (for any pov, except Potlatch itself) this was a valid concern. I understand you're caught between a rock and a hard place since transactions do not exist yet in the API, so there's isn't much you can do now to fix this. > by me, that database freakiness _will_ happen because we don't have > transactions, our server is often under high load, and there may be memory > leaks or blocking processes in some of the software we use. I don't mind you > getting annoyed because data isn't how you expect but I do object to you > blaming one particular component when it actually ain't that simple. So, AIUI, even JOSM or any other sw that works with relatively high loads of data at a time could bring up the same behaviour? -- Regards, EddyP ============================================= "Imagination is more important than knowledge" A.Einstein
signature.asc
Description: OpenPGP digital signature
_______________________________________________ dev mailing list dev@openstreetmap.org http://lists.openstreetmap.org/listinfo/dev