[ClientDisconnect] Un-hiding loot (alt key) causes client disconnect id#3,996,835,957

Hi!

I've invested significant time and effort to grow my character to be able to run "juiced" up challenging high-tier maps (delirium + triple beyond + nemesis+alva etc). However, I've found the experience to be "beyond" frustrating due to what appears to be PoE server scaling or client bugs.

Main symptom / TL;DR:

If I hit the alt key during a map with too many items, the client crashes back to the login. This kills a portal and totally sucks. Please look into this and fix it, or provide viable workarounds, so I can be less frustrated and better enjoy endgame mapping.

Debug details, background, observations+speculation:

High risk is supposed to be met with high reward: not a specialized class of scaling bugs. Often when pairing down what loot I'll be able to extract from a map (given a limited number of bag slots + portals) I'll use the alt key to verify the item level of items. This most often comes into play when deciding which amounts of divination cards, scarabs, or fragments to keep along with influenced items. I'll also use the alt key to gauge the tier/range of certain mods.. or trigger the alt key in the ctrl-alt command to link items to people. When I do this on very juiced maps with lots of loot on the ground the game client will pause and then kick me back to a login screen, wasting a portal on these high-volume maps, which are generally very expensive to craft and maintain. If I'm careful, I can locate a corner of the map where I haven't killed anything to use the alt key and review items in my inventory.

Obviously I couldn't create a debug before the crash, so I'm not sure if it's ultimately helpful to have a debug report after the fact -- but I imagine this should be trivial to reproduce for devs, by seeding a map with a significant amount of items on the ground in a single area [debug-1].

The last few lines of my client.txt log before/after the kick back to login are included [client.txt]. I've zipped up my loot filter(s) and productionConfig.ini (minus the login info) and shared it on dropbox [config]. Also, you can view what my client saw in this video clip [video.twitch]. The latency skyrockets and the fps drops to zero; whatever is happening seems to cause the server<-->client to become overwhelmed/non-responsive and die. Maybe this is the rendering of the items? You tell me.

Over time I've noticed the game client struggles to handle situations with lots of particles and large numbers of mobs or applying actions to large number of mobs. I suspect, this could be due to a high rate of context switching for single-core operations: given some very loose windows debugging, it appears that the code is burdened by client communication (interrupts) even with very beefy hardware. If there are functions that are trying to use multi-core operations, you may want to evaluate the cost of those. As the number or size of calls scale upward: multi-core operations incur significant latency penalties from the memory copy/context switch across logical processors. I suspect there are probably a large number of inefficiencies being triggered in the encryption/decryption calls and/or network code (a lot of people observe this as the "Legion effect", causing the client's code to become blocking for other operations (communication, rendering, input). I would suggest looking at how you're calling the network stack, whether it's a mutex or select (eventing? forking? hopefully not serial..) type of operation as that will most certainly lead to queueing (and then blocking) for this type of code/app. Total speculation, but this might be why folks end up in a "desync" state, on some fights (SIRUS): the client enters a blocking state, even if for a moment, for it's own on-going calls (clones, lazer-pinwheel, storms, ground-flares..). For this reason I actively avoid "more gore" cosmetic adding items and fancy MTX items (it's a shame I can't use many of the assets that I've paid for, without game impact).

I've observed that especially when trying to stream the game with OBS, the game is more likely to crash to the desktop; yet still haven't narrowed down if it's more related to render interface conflicts or maybe network stack priority. I suspect the latter, as I do not experience the same frequency of issues when just recording to my local system (vs streaming). Maybe this is due to inefficiencies with the anti-cheat system..? I don't know. I've gone through significant efforts of trial and error debugging to get my client to a mostly stable state at 1440p/144fps. Due to recent more-frequent observations of FPS drops and client<-->server frame desync since the latest patch, I've tried moving from 1440p/144fps (which again, has been mostly stable) to 1080p/120fps, in-case maybe that larger rendering/more frames means more network calls between client<-->server. Included with this change to a lower resolution, I wiped out all of my PoE asset caching directories to start fresh + reduce hash/indexing latency crawling over so many assets (another code inefficiency, easily observable with general filesystem syscall debug tools).

For reference, here are my system details (not a potato):

i9-9900K @ (8c/16t) @5GHZ
32GB 14ns DDR4
2TB m.2 NVMe
11G RTX 2080 Ti
ASUS ROG PG27UQ UHD/144hz | (PoE {was@2560x1440/144hz, {last 2 days} 1980x1024/120hz)

Network wise, I have a gigabit upstream connection and I maintain and monitor my local network as well as client health on my network very closely. I am keenly aware of the state of latency, loss, throughput and other health as well as performance parameters both in my own network as well as how things look from the outside in. I can make the data-demonstratable claim: this isn't an (observable) problem with [my network.jpg].

Thanks.

[debug-1](debug 3,996,835,957, taken after logging back in)
2020/08/20 21:05:48 28455546 b80 [INFO Client 4400] : Submitting bug report. Please wait for your bug report number to appear in chat, then take a screenshot (F8) and post in our bug report forum.
2020/08/20 21:05:48 28455593 b80 [INFO Client 4400] : Your bug report reference number is 3,996,835,957.


[client.txt]
2020/08/20 21:03:30 28316765 b80 [INFO Client 4400] Alva, Master Explorer: Time to go.
{global chat line removed for respect of privacy}
{global chat line removed for respect of privacy}
2020/08/20 21:03:53 28339812 15b [WARN Client 4400] Matching object found for InstanceClientActionUpdate, but no matching action was found on object to upda
te.
2020/08/20 21:03:53 28340531 15b [WARN Client 4400] Matching object found for InstanceClientActionUpdate, but no matching action was found on object to upda
te.
2020/08/20 21:03:53 28340531 15b [WARN Client 4400] Matching object found for InstanceClientActionUpdate, but no matching action was found on object to upda
te.
2020/08/20 21:03:53 28340656 15b [WARN Client 4400] Matching object found for InstanceClientActionUpdate, but no matching action was found on object to upda
te.
{global chat line removed for respect of privacy}
{global chat line removed for respect of privacy}
{global chat line removed for respect of privacy}
{global chat line removed for respect of privacy}
2020/08/20 21:04:28 28374937 15b [WARN Client 4400] Matching object found for InstanceClientActionUpdate, but no matching action was found on object to update.
2020/08/20 21:04:38 28385000 b80 [INFO Client 4400] Alva, Master Explorer: Good job, exile.
2020/08/20 21:05:03 28410609 899 [INFO Client 4400] Abnormal disconnect: An unexpected disconnection occurred.
2020/08/20 21:05:06 28413296 b7 [INFO Client 4400] Async connecting to sjc01.login.pathofexile.com:20481
2020/08/20 21:05:06 28413328 de [INFO Client 4400] Connected to sjc01.login.pathofexile.com in 0ms.
2020/08/20 21:05:09 28415906 e3b [DEBUG Client 4400] Got Instance Details from login server
2020/08/20 21:05:09 28415906 e58 [INFO Client 4400] Just before calling client instance session
2020/08/20 21:05:09 28415906 103 [INFO Client 4400] Connecting to instance server at 173.237.38.52:6112
2020/08/20 21:05:09 28415953 1bd [DEBUG Client 4400] Connect time to instance server was 31ms
2020/08/20 21:05:10 28417031 6ca [DEBUG Client 4400] ALTERNATE TREE JEWEL: Applying modifications for jewel with alternate tree version = 2, seed = 12049, and keystone selection = 3
2020/08/20 21:05:10 28417593 b80 [INFO Client 4400] : You have entered Karui Shores.


[config] (+filter{s})
loot filter + production config.ini (with username removed)
https://preview.tinyurl.com/harvesticlev4cnf


[video.twitch] (demonstrating holding alt, you can see latency spike up + fps drop to 0)
https://www.twitch.tv/videos/717234826

[my network.jpg]
https://i.imgur.com/8DGceCn.jpg

"

viable workarounds

1. Use very strict filter, disable "Key pickup", never press "Highlight items" key.

2. Modify your main filter and disable all SetFontSize entries as well as set all transparency to 255.

3. Make an "Angry reddit thread" related to this problem.

I don't have key pickup (and a few other keys) bound. I'll need to be sure to disable the highlight key when running these lootsplosion maps, which is unfortunate. Thank you for the tip on fontsize/transparency. Will incorporate that as I'm working through editing yet-another uber-strict-plus-plus-uber-alpha-beta-omega-hfs-wtf-bbq filter.

"

..by increasing the single thread performance of your CPU by 10 times :P

ez: "Patch 3.12: The Hadoop League"