Timelords
Checking Timelord Logs
Make sure to set your logs to INFO for the below commands to work.
To do so run: chia configure --log-level INFO
The below commands are for use with linux OS.
The tail
command only pulls new information as it is added to the log files showing new information as it is added.
The cat
command pulls all current data from the log files providing a snapshot of where it was and is currently.
The matched strings are missing the first letter (E or S) intentionally.
This is due to the two instances of each log, for example the log where others create the proof of time has the string "Not skipping" while the log where your machine creates the PoT has the string "Skipping".
Continuous:
tail -F ~/.chia/mainnet/log/debug.log | grep "stimated"
: running output of the current estimated IPS as seen by the network (note this will be lower than the IPS reported by the ASIC software).
tail -F ~/.chia/mainnet/log/debug.log | grep "kipping"
: running output of peak heights being added to the node. If a height is skipped than very likely your timelord created the PoT, if the height is not skipped then a different timelord created the PoT.
Static:
cat ~/.chia/mainnet/log/debug.log | grep "stimated"
: static output of the current estimated IPS as seen by the network (note this will be lower than the IPS reported by the ASIC software).
cat ~/.chia/mainnet/log/debug.log | grep "kipping"
: statis output of peak heights added to the node. If a height is skipped than very likely your timelord created the PoT, if the height is not skipped then a different timelord created the PoT.
Example Estimated IPS Logs
These logs indicate your timelord has completed a Proof of Time, note there are three VDF chains and the specific VDF that was completed will be indicated. Also note that this does not mean your timelord is the fastest timelord, only that it completed a PoT.
cat ~/.chia/mainnet/log/debug.log | grep "stimated"
Response:
2024-04-11T16:59:01.482 timelord chia.timelord.timelord : INFO Finished PoT chall:79a8b30ee7aaa3064ac4.. 1981176 iters, Estimated IPS: 575336.2, Chain: Chain.REWARD_CHAIN
2024-04-11T16:59:01.588 timelord chia.timelord.timelord : INFO Finished PoT chall:027e1c04ae11d627179b.. 1981176 iters, Estimated IPS: 558173.6, Chain: Chain.INFUSED_CHALLENGE_CHAIN
2024-04-11T16:59:01.627 timelord chia.timelord.timelord : INFO Finished PoT chall:50cd37772d5bae2a00d4.. 1981176 iters, Estimated IPS: 552043.3, Chain: Chain.CHALLENGE_CHAIN
Example Skipping Peak Log
These logs indicate your timelord is skipping the peak height, one can reasonably assume that they have completed the Proof of Time the fastest or at least as fast as the other timelords.
Note: a capital S
is used here to just pull the Skipping peak logs.
cat ~/.chia/mainnet/log/debug.log | grep "Skipping"
Response:
2024-05-01T06:30:27.983 timelord chia.timelord.timelord_api: INFO Skipping peak, already have.
Example Not Skipping Peak Log
These logs indicate your timelord is not skipping the peak height, this means that your timelord was not the fastest at completing the PoT and it is getting the peak from a peer node.
Note: a lower case s
is used here to just pull the Not skipping peak logs.
cat ~/.chia/mainnet/log/debug.log | grep "skipping"
Response:
2024-05-01T06:30:27.983 timelord chia.timelord.timelord_api: INFO Not skipping peak, don't have. Maybe we are not the fastest timelord.
If you have set up the timelord services to run through systemD you can use the below command(s) for pulling those records.
If running as a service:
journalctl -f -u chia-hw-vdf | grep "MHz"
: output of the ASIC timelord reported IPS (this will be higher than the estimated IPS as seen by the network). If not using systemD this information is available in the the terminal output where the hw_vdf_client
command was run.
Common Timelord Error Logs
Chia logs
timelord chia.timelord.timelord : ERROR Error while handling message: Traceback (most recent call last):
File "chia/timelord/timelord.py", line 893, in _manage_chains
File "chia/timelord/timelord.py", line 440, in _submit_iterations
File "asyncio/streams.py", line 359, in drain
File "chia/timelord/timelord.py", line 221, in _stop_chain
This error occurs when the ASIC software has been shutdown unexpectedly and will start spamming your log file.
The current resolution is to restart the chia processes with something like chia start full_node timelord-only -r
.
ASIC logs
Warning: too much work for VDF aux threads
- This generally means that the frequency is set too high, it is recommended to use the auto-frequency flag for the software to self adjust and resolve this error --auto-freq 60
updates the frequency every 60 seconds.
Bad data size after stop: -1
- This is a known logging error related to how the chia client and ASIC disconnect after a stop. There is no concern with having this error in your logs.
Timelord support
Join Our Discord and jump into the #support channel for support