#saltr (irc2p) | I2P+ 2.5.2+ >> i2pplus.github.io | skank.i2p | git.skank.i2p | purokishi.i2p | vituperative.github.io/i2pchat | natter.i2p | notbob.i2p | ramble.i2p | shreddit.i2p | cake.i2p | /msg an op for voice

y2kboy23 dr|z3d snark refresh is broken. Seeing "Uncaught TypeError: filterbar is null" in the dev console.

y2kboy23 well, once I enabled it, it started refreshing

T3s|4 y2kboy23: do you mean once I enabled the 'Torrent filter bar'?

dr|z3d y2kboy23: thanks, will fix that.

dr|z3d on the plus side, no pun intended, you got exposed to a feature you obviously weren't using :)

dr|z3d T3s|4: have a look at snark's configs, you'll see what y2kboy23 means.

y2kboy23 Sorry. With the filter bar off, the refresh is broken.

zzz dr|z3d, I think you're in desperate need of concurrency help?

dr|z3d if you're offering, zzz, I'll take any help I can get, thanks. it's what I'm working on right now.

zzz yup

dr|z3d synchronized isn't cutting it. I was going to move the removal of stale entries inside the writer, but haven't got there yet.

zzz first of all, I recommend the concurrency section of Thinking In Java

zzz but here's a tl;dr

dr|z3d concurrentHashMap I tried, no dice there either.

zzz when using non-thread-safe data structures such as HashMap, ArrayList, etc., there's only 3 options to avoid ConcurrentModificationExceptions and weird bugs

zzz 1) ensure ALL accesses (get(), put(), remove(), clear(), etc) are from a single thread

zzz 2) synchronize around ALL accesses with the same lock

zzz 3) switch to a thread-safe-datastructure such as ConcurrentHashMap

zzz there's an additional requirement when removing an element inside a loop

zzz let me look at your takes, just a sec

dr|z3d you've summarized nicely what I've discovered. I think I was/am probably omitting the file lock element with synchronized which is why it's not working as intended.

zzz no. you need to synch at _every_single_place_ you touch rdnscache

zzz and yes, to remove inside a loop, you must use iter.remove(), and you must never add within a loop if you're looping over that thing

dr|z3d yeah, got that. my local copy has that implemented.

dr|z3d the synchronized everywhere part.

zzz ConcurrentHashMap should also work, subject to the loop requirements above

zzz that's it, have fun

dr|z3d thanks, appreciate the insights.

dr|z3d this is the current iteration of the relevant methods, fwiw. cake.i2p/view/nzTmK9ETeA_N28jAPcD81cTS1ZRg8dr2c2UJh2B0W_OpEmw4X4aU/nzTmK9ETeA.txt

dr|z3d I noticed you got the rough end of Opicaak earlier. Ugly :|

zzz you can see why java 5's ConcurrentHashMap was a godsend

dr|z3d Yeah, I obviously didn't do it right when I tried a ConcurrentHashMap, it got rid of the errors but also didn't bother to read in the filecache to memory.

dr|z3d I'm going to add a couple more synchronized to that code block I just caked and see what gives.

dr|z3d the error took a while to manifest because stale entries in the file cache are only removed when the # entries exceeds the max configured size, 32K in most cases.

zzz you also need to synch readRDNSCacheFromFile, and lots of spots that are not in that pastebin

dr|z3d yup, done that, testing code.

dr|z3d what's your appraisal of the general idea? nuts? or? :)

zzz mildly nuts

dr|z3d :)

dr|z3d It makes for a) a much more informative console when looking at tables and the netdb.

dr|z3d and b) with a file cache, zero delay once the cache is populated and minimal lookups.

zzz one other subtlety: public synchronized static foo() is a DIFFERENT lock than public synchronized bar()

dr|z3d so all methods being synchronized on the same file need to be either static, or not.

zzz yup

dr|z3d ok, good to know, thanks. I'm all static for the file cache methods.

zzz the former is the same as synchronized(BlahBlah.class) { ... } and the latter is the same as synchronized(this) { ... }

zzz the best practice would be to do neither, but to synch on the object in question, i.e. synchronized(rdnsCache) { ... } everywhere

zzz but your way will work

dr|z3d current revision, will take a few minutes for errors to generate, or not: cake.i2p/view/DlyObgeFeT_j54qTDyqRWx3EtCyrr8RLhpXFAPckk_jZUw0X25KO/DlyObgeFeT.txt

dr|z3d a fully populated 32K entry file cache is ~1MB, which isn't too bad.

dr|z3d as I thought, still the same error.

dr|z3d this method:

dr|z3d private static synchronized void cleanupRDNSCache() {

dr|z3d List<String> keysToRemove = new ArrayList<>();

dr|z3d Iterator<Map.Entry<String, CacheEntry>> it = rdnsCache.entrySet().iterator();

dr|z3d while (it.hasNext()) {

dr|z3d Map.Entry<String, CacheEntry> entry = it.next();

dr|z3d if (rdnsCache.size() > MAX_RDNS_CACHE_SIZE) {

dr|z3d keysToRemove.add(entry.getKey());

dr|z3d it.remove();

dr|z3d }

dr|z3d rdnsCache.keySet().removeAll(keysToRemove);

dr|z3d }

dr|z3d really does not like Map.Entry<String, CacheEntry> entry = it.next()

zzz well, for one, you either need it.remove() or the whole keysToRemove thing, dont do both

zzz it.remove() is a lot more efficient

dr|z3d ok, I'll try that, thanks.

zzz but that doesn't explain why it's not working

T3s|4 dr|z3d: yep, on the snarkified 'plus' side, seems to have been working / is working as expected :)

zzz did you sync the methods below cleanupRDNSCache() that you did not include in the paste?

dr|z3d thanks for reporting back, T3s|4. y2kboy23's issue was easily fixed, I put something in last minute without testing the badge refresh properly.

T3s|4 great

dr|z3d zzz: this one, yes, but I have a feeling there's another one that needs the syncrhonized treatment:

dr|z3d public static synchronized int countRdnsCacheEntries() {

dr|z3d return rdnsCache.size();

dr|z3d }

dr|z3d like probably this one: public String getCanonicalHostName(String ipAddress) {

zzz bingo

dr|z3d although not sure about that, and it's not a static method.

dr|z3d let's see if it moans if it's converted to static :)

zzz you'll have to take out the logging

dr|z3d yeah, I was using system.err where the logging wasn't working.

dr|z3d will remove it once the whole thing is working properly, don't want to be polluting wrapper logs.

zzz FYI, java overhead for a Map entry is about 120 bytes, and the way you did it is even more, so your size estimate is at least 4X low

dr|z3d talking about the on-disk cache.

dr|z3d I figured the in-mem cache would be larger, 4x filecache, not bad. 4MB when it's full on a >=512M router, acceptable.

zzz why in the world is the map IP->CacheEntry(IP, hostname) rather than just IP->hostname ??? Why store the key inside the value?

zzz that's adding about another 56 bytes per entry

dr|z3d that's a good question. :)

dr|z3d so I had to implement a lock object to bridge the static and non-static methods.

dr|z3d testing code, let's see if it works.

dr|z3d so it looks like this might be working, thanks for the help, zzz. I'll have a crack at the cache entry storage next. cake.i2p/view/1np26pHE10_RGppfzDvzYLxRIyE76Z6QMKwagrzv3_CPRrXEIH6b/1np26pHE10.txt

zzz ip->hostname would disallow null hostnames; not sure if you want to store those as a kind of negative lookup cache mechanism

zzz you'd have to switch to a separate negative lookup cache, or use something else like "" to indicate you already looked it up

dr|z3d ok, what about this? cake.i2p/view/Ru3HYf2KEC_WSQey62aX0ewHout1M0PFmPhzK0YsC_NNbeUkYvrk/Ru3HYf2KEC.txt

zzz ja, though as I said if you're doing that, do the best practice of syncing on rdnsCache itself (and make it final), no need for a separate lock

zzz but your way works fine

dr|z3d that latest cake should address the ip/hostname issue in the mem cache.

dr|z3d the lock stuff is working a.ok, no more concurrent mod errors in console.

dr|z3d noted re syncing on the cache, thanks.

zzz no, my suggestion is Map<String, String> rdnsCache, making it a map of IP -> hostname, and nuking the CacheEntry class completely

zzz this would vastly simplify it but would be a big refactor

zzz there should never be a need to store the key inside the value, that's what a Map does for you, it associates a key to a value

dr|z3d yeah, right, so I've removed that, the ip is now the key.

zzz ok, you're fast, but it's not in the last cake above

dr|z3d oh, then I must be doing it wrong. I thought: rdnsCache.put(cacheEntry.getIpAddress(), cacheEntry); would address it.

dr|z3d more work to do, thanks. will ping you again when I'm sure I've fixed it, though possibly not with a total refactor just yet.

zzz I'm saying you don't need class CacheEntry at all. rdnsCache.put(ip, hostname); hostname = rdnsCache.get(ip)

zzz but yes it's a big change to something you just got working, wouldn't blame you if you didn't, it's just not great code

zzz I'm saying you don't need class CacheEntry at all. rdnsCache.put(ip, hostname); hostname = rdnsCache.get(ip)

zzz but yes it's a big change to something you just got working, wouldn't blame you if you didn't, it's just not great code

zzz the map "keeps" the key and value "together". You don't need to put them both in a CacheEntry class to do that

dr|z3d hey, it works, we're not all java sexperts like you :)

dr|z3d but thanks for the feedback, will look at a refactor maybe in the next release cycle.

Irc2PGuest60863 it might be useful to have a CacheEntry if you want to keep timestamp with the hostname

Irc2PGuest60863 (which is usually how caches work, remove the oldest or lru)

Irc2PGuest60863 (also pretty sure jdk has a built in lru cache)

dr|z3d I don't want a timestamp with the hostname, just makes the cache bigger. We're storing the cache entries oldest first, so they get removed when the cache is full.

Irc2PGuest60863 Blinded message

Irc2PGuest60863 what makes you think you're storing the oldest first?

Irc2PGuest60863 though admittedly I just glanced at the code

zzz right

zzz a HashMap don't do that

Irc2PGuest60863 I could've sworn java has a built in class that does this

zzz LinkedHashMap

Irc2PGuest60863 yeah LinkedHashMap gets you 90% of the way there I think

dr|z3d patches welcome, as is the common refrain :)

Irc2PGuest60863 what version of java do you use anyways? is the router still restricted to java8?

zzz our LHMCache class gets you the other 10%

dr|z3d Java8 is what we're currently requiring.

dr|z3d LinkedHashMap rework: cake.i2p/view/OeWflKR7pq_vHxjlY4nMau8ENK1FKYdiSyBYaw3aL_WpEqsyDjBf/OeWflKR7pq.txt

dr|z3d I feel like you want to pull this code into canon, zzz :)

Irc2PGuest60863 how do you retype the code so fast hehe

zzz we have a no nuts policy in canon

dr|z3d you need to re-evaluate the code. this is lovely, ripe fruit :)

Irc2PGuest60863 dr|z3d: I worry about that writeRNDSCacheToFile method which calls isDuplicateEntry on every line which re-reads the file?

Irc2PGuest60863 but the code seems fine. If I was really reviewing the code I might say it's not... OO enough. But some people don't like my OO style. Also obtaining a lock to write stuff to disk is rarely necessary

Irc2PGuest60863 in java 99% of the time you're better off making a copy of a data structure and then writing that then synchronizing

Irc2PGuest60863 then again the cost of synchronization is so low these days eh

Irc2PGuest60863 I'll also always wonder about the use of statics within the router codebase

Irc2PGuest60863 none of that code needs to be static afaict

dr|z3d fire up git, pull the code, see if you can improve it.

Irc2PGuest60863 hehe he said fire up git

Irc2PGuest60863 git's an exe

dr|z3d I'll push the latest revision once I've determined it's not hosing stuff.

Irc2PGuest60863 dr|z3d: I'll improve it if you can tell me why I haven't been able to load postman all day heh

dr|z3d what version are you running?

dr|z3d of i2p+?

Irc2PGuest60863 I'm actually genuinely curious what's going on. Before I thought it was a matter of router integration but it's been 27 hours

dr|z3d 58+ perchance?

Irc2PGuest60863 dr|z3d: yeah I'm running the latest version of i2p+ as usual. everything else seems to be working fine... except postman

dr|z3d is that 58+ or something more recent?

Irc2PGuest60863 62+

dr|z3d because 65+ is current right now.

Irc2PGuest60863 really I guess I could upgrade again. you think it might help

dr|z3d regarding postman, I don't know. I'd wait a few if you are going to upgrade, I'll push a new update with the RDNS changes once I'm confident they're working as intended.

Irc2PGuest60863 k

dr|z3d I'm reverting the linkedHashMap and key/value changes for now, not working as intended, file cache is getting overwritten. Need to look at that again later.

dr|z3d ok, -66+ uploaded.

zzz postman working for me on canon trunk

Irc2PGuest60863 dr|z3d: my code would probably look something like cake.i2p/view/kTjN0xzM5L_fE4jYPdWorY2UN4LyPsrURKWMtBJMr_deminNcWcT/kTjN0xzM5L.txt

Irc2PGuest60863 I will note that (1) you don't want to copy the tmp file and then delete it. Instead you want to use an atomic file move. That code as is is buggy and will break because copy isn't atomic.

Irc2PGuest60863 (2) you don't need t lock the map to write it to disk, just create a copy. Map<String,String> copyToWriteToDisk = new HashMap<>(ipToHostname) does the trick

Irc2PGuest60863 (3) I still don't understand what isDuplicateentry is doing in that code. it's pretty strange.

Irc2PGuest60863 lastly I'm never a fan of developers creating their own adhoc file formats, hehe. just use xml. But that dear friends is a story for another time

dr|z3d ip,hostname

dr|z3d it's pretty simple, keeps the file as small as possible, and is human readable.

Irc2PGuest60863 yeah probably, in this case, I'm just never a fan of people inventing file formats

dr|z3d it's .txt, it's not a new format.

Irc2PGuest60863 there's strange edge case you haven't thought of that would be handled by an existing file format

Irc2PGuest60863 even using a Java Properties Writer might be... safer

Irc2PGuest60863 also I might keep the Cache Entry class. The reason is you probably want to remember when you did the resolution and got the hostname. During initialization when you read in the file I would ignore any cache entries older than 30 days

dr|z3d as for isDuplicateEntry, it's meant to check for dupes.. the whole point of a simple format is to allow you to manually edit it, merge files if required.

Irc2PGuest60863 stuff like that happens when somebody installs i2p, stops running it, then comes back 2 years later and restarts the router and now the info in your cache might be wrong

Irc2PGuest60863 dr|z3d: why do you need to check for dupes in a map?

dr|z3d there's sufficient churn to not need to care about the timestamp.

dr|z3d you don't, you need to check for dupes in the file.

Irc2PGuest60863 dr|z3d: but if the only thing you ever write to the file is a map how could dupes get in the file?

dr|z3d I just told you.

dr|z3d you might manually merge files.

Irc2PGuest60863 hehe "sufficient churn to not need to care about the timestamp" == famous last words

Irc2PGuest60863 dr|z3d: when you read the file into a map in memory dupes will be eliminated

Irc2PGuest60863 when you write the file to disk there can be no dupes

Irc2PGuest60863 so all that code has to do is make sure to read the file and then write the file and use the ATOMIC_MOVE option not the copy operation

Irc2PGuest60863 and yes you've invented a new file format. just understand that's a risky thing to do. most file formats in i2p for example are accompanied by a formal spec. but yeah whatever

dr|z3d open file, see format. :)

dr|z3d which is the point. human readable. I don't want xml or anything else which obfuscates the information.

Irc2PGuest60863 xml is much more human readable than a custom format

dr|z3d actually, no. not in this case.

Irc2PGuest60863 but I digress, my main point was that you shouldn't need to check for duplicates

Irc2PGuest60863 and a timestamp isn't a crazy idea

dr|z3d ok, thanks, noted. I'll put that theory to the test later.

dr|z3d I initially had timestamps, just found them to be unnecessary, first in, first out is how the cache _should_ function.

Irc2PGuest60863 I'd also suggest that storing cached data in a config directory is strange. It's at best tmp data. the sort of thing you dont' really want to keep if you for example switch router versions. not sure the router policy on this but it's usually a good idea to distinguish between configuration data and runtime/"workspace" data and "tmp" data

dr|z3d no, it's definitely meant to be in .i2p/

Irc2PGuest60863 dr|z3d: timestamps aren't for eviction. they're a form of sanity checking. if you read in cache data which is, say, 2 years old, then it should all be thrown away

Irc2PGuest60863 you're assuming ppl are regularly restarting their routers... which is an assumption that won't always hold

dr|z3d first in, first out. like I said, there's enough churn to remove stale entries soon enough. the hostname data isn't critical.

dr|z3d no assumption relating to router uptime.

dr|z3d We just don't want to store cache data in a volatile location, it's in .i2p because it needs to be persistent.

zzz 'dupes impossible in a map' is not a theory

zzz 'why put the key on the value side of a key/value store' is neither a Java question nor a programming question

zzz I doubt the file is in ~/.i2p but if it is it's not on purpose

zzz the file format objection is silly but if you claim it's a csv file you might win the argument

zzz we have his ATOMIC_MOVE in our utils as FileUtil.rename()

zzz "xml is much more human-readable than csv" is an insane take

dr|z3d did I miss part of this conversation. not sure.

dr|z3d happy with the file format, that's not going to change.

zzz no, just comments on the last half of the rdns discussion where ircguestxxx got involved

dr|z3d location for the file is ~/.i2p/ in purpose.

dr|z3d *on purpose

zzz do me a favor and look if it's there

zzz on your disk

dr|z3d it is.

dr|z3d always has been. by design.

zzz not ~/i2p/ ?

dr|z3d no

zzz ah never mind, I see where you did it

zzz missed it because it's not how I would have done it

zzz see, told you it wouldn't be long before I was wrong

dr|z3d :)

zzz ofc ~/.i2p is the right place, as he said he doesn't know what the 'router policy' is

dr|z3d you're right about xml, you'd have to be smoking crack to suggest that was more human readable than what I've implemented :)

zzz no, not just more -- "much more"

zzz you also presumably yanked that code from one of the 20 times I've done it, so I'll take your side on that one

dr|z3d right. and /tmp/ is definitely the wrong place.

dr|z3d I forget, but probably :)

dr|z3d the main thing right now is that it works and doesn't cause console errors. as you've pointed out, it could be made much more efficient and I'll have another pass at it post release, probably.

dr|z3d it's the difference between waiting 30s or more for one of the pages containing large peer tables to load and having the page load almost instantly. so I consider it a win, all things considered.

dr|z3d oh, and if it's not obvious, ircguestxxx is mesh.

dr|z3d who still hasn't reported back on the update notifications he specifically asked for.