map load lag: proposal for some new settings #3299

New Issue

flux · 2022-12-22T17:42:30Z

flux commented

2022-12-22 17:42:30 +00:00

currently, when the server has about 29+ players, loading new mapblocks can take forever. i've got a possible idea of how to improve performance by changing some settings.

emergequeue_limit_total = 10752  # = 42 * 256; default 1024
num_emerge_threads      = 0      # default 1

it seems that having num_emerge_threads = 1 doesn't prevent #850 from occurring, though it possibly reduces its frequency. however, i've got a theory that this is the reason that loading the map takes so long when the server is otherwise not over-burdened. increasing emergequeue_limit_total will also ensure that some clients aren't totally starved out of asking for new blocks to load.

because testing this requires actual users doing actual things, i don't think this is something we can test effectively on the test server.

currently, when the server has about 29+ players, loading new mapblocks can take *forever*. i've got a possible idea of how to improve performance by changing some settings. ``` emergequeue_limit_total = 10752 # = 42 * 256; default 1024 num_emerge_threads = 0 # default 1 ``` it seems that having `num_emerge_threads = 1` doesn't prevent #850 from occurring, though it possibly reduces its frequency. however, i've got a theory that this is the reason that loading the map takes so long when the server is otherwise not over-burdened. increasing `emergequeue_limit_total` will also ensure that some clients aren't totally starved out of asking for new blocks to load. because testing this requires actual users doing actual things, i don't think this is something we can test effectively on the test server.

flux added the

1. kind/balancing

1. kind/other

labels 2022-12-22 17:42:30 +00:00

AliasAlreadyTaken was assigned by flux

2022-12-22 17:42:30 +00:00

flux self-assigned this 2022-12-22 17:42:30 +00:00

AliasAlreadyTaken commented

2022-12-23 14:47:03 +00:00

Changing those values is not too dangerous. Worst that can happen is we see more artifacts and cutoff treetops.

AliasAlreadyTaken referenced this issue

2022-12-23 14:49:14 +00:00

1.1.117 #3128

flux referenced this issue

2023-01-13 23:45:09 +00:00

mrminer reports: such lag chunk load so slow ... #2653

AliasAlreadyTaken commented

2023-01-14 04:48:35 +00:00

Those values are online for more than a week now, with little impact on the maploading or database lag.

👍 1

flux commented

2023-01-14 20:33:49 +00:00

i guess we should close this then? we can either revert the values, or not, it seems they don't matter? we need more data on where the bottleneck is.

👍 1

AliasAlreadyTaken closed this issue

2023-01-15 00:40:20 +00:00

AliasAlreadyTaken added the

5. result/fixed

label 2023-01-15 00:40:36 +00:00

AliasAlreadyTaken commented

2023-01-15 00:42:39 +00:00

Although those values did not "fix" the issue at hand: the map_saving_interval lag

However, the values were looked at, so the request to test them is resolved.

Although those values did not "fix" the issue at hand: the map_saving_interval lag However, the values were looked at, so the request to test them is resolved.

flux commented

2023-01-26 00:09:21 +00:00

actually, not sure if something else changed, but i felt like the map was quite responsive for me today even w/ 33-38 players.

mesecons_debug lag got up to 40 too - it seems like the mapgen was slowing everything else down by a large but relatively stable amount - there weren't big lag spikes.

actually, not sure if something else changed, but i felt like the map was quite responsive for me today even w/ 33-38 players. mesecons_debug lag got up to *40* too - it seems like the mapgen was slowing everything else down by a large but relatively stable amount - there weren't big lag spikes.

Sign in to join this conversation.