#i2p-dev (irc2p) | 2.8.2-6

zzz but yeah, I'm out too, later...

eyedeekay I was trying to figure out how to package docs together with the code, each directory has a README.md that contains the godoc and that call tree formatted

eyedeekay \/tree/graph/

RN net.i2p.router.JobQueueRunner : Error processing job [Handle Database Lookup Message] on thread 1: java.lang.ArrayIndexOutOfBoundsException: arraycopy: length -8 is negative

RN <Afkaid> java.lang.ArrayIndexOutOfBoundsException: arraycopy: length -8 is negative

RN anything of concern?

RN taking an overdue look at my logs, and I have a crit with a negative length:

RN CRIT [uildExecutor] uter.tunnel.pool.BuildExecutor: B0rked in the tunnel builder

RN java.lang.ArrayIndexOutOfBoundsException: arraycopy: length -9 is negative

RN both mention arraycopy

RN can add ntcp reader to the list

RN doubt the value of the negative value matters but I can get the negative if it will add to a positive outcome

RN </math_nerd_joke>

orignal zzz, yes, I'm doing the same changes

dr|z3d cannon or canon, RN?

Afkaid CRIT [uildExecutor] uter.tunnel.pool.BuildExecutor: B0rked in the tunnel builder

Afkaid java.lang.ArrayIndexOutOfBoundsException: arraycopy: length -25 is negative

dr|z3d is that normal I2P or I2P+, Afkaid?

Afkaid normal

Afkaid 2.8.1-0

dr|z3d if you've got a stack trace with the error, you might want to paste it on cake.i2p with a 24h expiry.

dr|z3d cake.i2p/pastebin

Afkaid cake.i2p/file/17KKGA3mgt_1W3opXPDkrdXIcR95jNyaaOpZTVk82_jW6aYiYKoG/I2P-2.8.1-0-CRIT-1.txt

Afkaid cake.i2p/file/Se0ArHdtvi_EDmacla906vFskzFOKgyYLIVYNUXv2_eqy0I1R9nM/I2P-2.8.1-0-CRIT-2.txt

dr|z3d looks like it relates to this method:

dr|z3d /**

dr|z3d * Calculate the hash and cache the result.

dr|z3d * @param source what to hash

dr|z3d */

dr|z3d public final Hash calculateHash(byte[] source, int start, int len) {

dr|z3d MessageDigest digest = acquire();

dr|z3d digest.update(source, start, len);

dr|z3d byte rv[] = digest.digest();

dr|z3d releaseit(digest);

dr|z3d return Hash.create(rv);

dr|z3d }

dr|z3d or maybe the issue's upstream with java.. what java version?

Afkaid Where do I find Java version?

dr|z3d on /logs

Afkaid 17.0.14

Afkaid The notification bubble on the left panel is nice to spot the critical errors :-)

dr|z3d :)

Afkaid good idea

dr|z3d haven't implemented that in + yet. on the todo list.

Afkaid I should try + one day

dr|z3d are the errors you're seeing frequent?

dr|z3d I just wonder if a java upgrade might be worth a try, 21 or 23 perhaps.

Afkaid Since the 2.8.1-0 update, I saw the first one 2 times, and the second one 1 time

dr|z3d ok, well, zzz will be around later and he'll have a look.

Afkaid I don't see any newer version on the repository

dr|z3d 24 is technically current now, but your repo may not be offering newer versions.

Afkaid yes

Afkaid these errors appears after I played a bit with the parameters, increasing the bandwidth values and forcing floodfill

dr|z3d probably not related.

Afkaid ok, I have to go, good luck with this one!

dr|z3d aight, laters o/

RN dr|z3d, the three errors were from Canon. IDK already ACK.

RN afkaid and I both saw them

zzz I saw one similar to Afkaid's but different. RN, need your full stack trace

zzz @@

zzz oh wow I'm getting a lot of them on one router

zzz hopefully the bubbles will help us catch things sooner

zzz seeing it about once an hour on a high-traffic router, so I should be able to narrow it down soon

zzz sadly, this will force a 2.8.2

dr|z3d possibly the same issue? cake.i2p/view/SjjArX0zjp_9pU5JLiVbLfkIC1nDqhXmzCk8djwZf_seyDcOSc4w/SjjArX0zjp.txt

dr|z3d this chunk of the stacktrace looks like it might be the root cause:

dr|z3d | INFO | Compiling: 21331 4 java.lang.invoke.MethodType::methodType (69 bytes)

dr|z3d | INFO | Compiling: 21324 ! 4 net.i2p.router.networkdb.kademlia.PersistentDataStore$ReadRouterJob::read (1423 bytes)

dr|z3d | INFO | - waiting on <no object reference available>

dr|z3d | INFO | - parking to wait for <0x000000072c7340a0> (a java.util.concurrent.SynchronousQueue$Transferer)

dr|z3d | INFO | - parking to wait for <0x00000006062f6d48> (a java.util.concurrent.SynchronousQueue$Transferer)

dr|z3d | INFO | - locked <0x000000072dab92f0> (a net.i2p.client.streaming.impl.PacketLocal)

dr|z3d | INFO | - locked <0x000000072d9cf3c0> (a java.lang.Object)

zzz no

dr|z3d different issue entirely then?

zzz but I've looked at your PersistentDataStore mods recently, and I think they're deeply flawed, and in the wrong place

dr|z3d only seeing this on a single router, haven't seen it elsewhere.

zzz because I looked at doing what I think you're trying to do, which is not write out some RIs, and I would do it a lot differently

dr|z3d ok, so at least conceptually we're on the same page, even if my implementation sucks :)

dr|z3d and yeah, I'm trying to keep "uniniteresting" RIs off storage and in RAM only.

zzz can't make heads or tails out of that paste but it seems to point to PersistentDataStore, maybe

dr|z3d smells like it, not sure what no object reference available points to, maybe OS level fail. dunno.

RN zzz, pm'd a cake link with some log snips... let me know if you neeed more :)

zzz I'm good, look similar to the ones I have, thanks

zzz drz, no time to do a full breakdown of your PDS mods but here's where I started as a simple alternative:

zzz @@ -1441,6 +1446,9 @@ public abstract class KademliaNetworkDatabaseFacade extends NetworkDatabaseFacad

zzz // + new Date(routerInfo.getPublished()));

zzz _context.peerManager().setCapabilities(key, routerInfo.getCapabilities());

zzz + // don't store old routers to disk

zzz + if (persist && VersionComparator.comp(routerInfo.getVersion(), StoreJob.MIN_STORE_VERSION) < 0)

zzz + persist = false;

zzz _ds.put(key, routerInfo, persist);

zzz if (rv == null)

zzz _kb.add(key);

zzz untested

zzz well, tested not to explode, but not verified that it works

dr|z3d ok, that looks vaguely similar to what I'm doing, in part at least.

zzz was it right after startup? because if it was compiling PDS it must have been...

dr|z3d I'm also not storing RIs to disk if they don't meet a b/w tier threshold iirc.

zzz it it was at startup, restart and chalk it up to JVM insanity

zzz I suggest you stare hard at PDS though to verify it's doing what you think it is

dr|z3d doesn't appear that the error occurs right after startup, no.

zzz and/or stare at your RI files on disk to verify it's excluding what you want it to

dr|z3d though it might have been, the errors were repeating. some memory management issues on that box, might be related.

dr|z3d I've started at those long enough to know it's doing what it should be. On startup, the netdb accurately reflects the filtered RIs.

dr|z3d *stared

zzz ok

zzz it seems like the JVM ran out of... something?

dr|z3d could be RAM, like I said, box is acting a bit strange, so probably something specific to that box.

zzz ok, since I'm waiting for a router to restart, I'll give you one more clue on why I think it's bad code in PDS

zzz in PDS.write() :

zzz ... calculate isOld, isBad, noSSU, isSlow, and a whole bunch of stuff....

dr|z3d yup

zzz if (dbFile.lastModified() < dataPublishDate) {

zzz ... write the file to disk ...

zzz } else {

zzz ... deal with all your isXXX stuff and take actions...

zzz }

zzz so all your added code is only if you did NOT write the file to disk???

zzz and, btw, that conditional should always be true, you're always writing new RIs to disk, so it seems like you would rarely hit all that code ever

dr|z3d yeah, ok, I think I see what you mean. I probably want to check the various conditions at the top, ie if (dbFile.lastModified() < dataPublishDate && !isBanned etc etc)

dr|z3d or just create a new boolean, isGood or something, and only write to disk if those conditions are met.

dr|z3d otherwise, spit out the debug logs as before.

dr|z3d thanks, I'll optimize and fix.

eyedeekay What are you guys doing to reproduce the CRIT? I've got a reachable box on the release build, forced floodfill, high participating traffic, 1GB RAM for the JVM, can't seem to make it happen naturally

zzz yeah drz it just seems wildly foobarred, and in the wrong place, put it in KNDF like my patch, putting policy down in PDS seems wrong

zzz eyedeekay, what JVM are you on?

eyedeekay 17 on that machine, with the Debian package

zzz I'm seeing a few per hour on the deb box which is Java 11, the other report was Java 17. Never seen on my boxes

eyedeekay Ah you're looking at the deb box. I had a similar idea the machine I'm talking about is the git box

zzz ok I just found the issue, will put up a -1-rc soon

eyedeekay OK I'll be ready to look it over, ping me

zzz it's related to using the SHA256's from the pool in Noise

zzz we're doing a double-free of them back to the pool, so then we have two threads using the same SHA256 at once

dr|z3d so needs a synchronized?

zzz yes and a already-freed check

zzz dunno why it's so common on some boxes and not others

zzz the pool change was on Feb. 7 so I'm a little stumped on why it wasn't seen until after the release

dr|z3d could be java version related? not seeing it at all here, I think I'm on at least 21 on all my boxes.

zzz shouldn't be, but who knows