Especially during initial history sync, there are a lot of False->False up_to_date transitions
(e.g. adb.add_address generates one), and the GUI does some work for each, which adds up to a lot
of CPU usage for the full sync.
- AddressSynchronizer no longer has its own state re up_to_date,
it defers to Synchronizer/Verifier instead
- Synchronizer is now tracking address sync states throughout their lifecycle:
and Synchronizer.is_up_to_date() checks all states
- Synchronizer.add_queue (internal) is removed as it was redundant
- should fix wallet.is_up_to_date flickering during sync due to race
related:
dc6c4814061c20a29a22
This reverts commit dc6c481406 as it introduced its own issue:
while add_address was running on one thread, synchronizer._reset could be running on another,
and by the time the "enqueue" coro would run, it would use a new add_queue and
addr would not be in requested_addrs anymore...
```
I/w | wallet.Standard_Wallet.[test_segwit_2] | starting taskgroup.
I | lnworker.LNWallet.[test_segwit_2] | starting taskgroup.
E/i | interface.[testnet.qtornado.com:51002] | Exception in run: KeyError('tb1q3wmgf8n5eettnj50pzgnfrrpdpjmwn37x7nzsc5780kk4je9v4hspym8mu')
Traceback (most recent call last):
File ".../electrum/electrum/util.py", line 1243, in wrapper
return await func(*args, **kwargs)
File ".../electrum/electrum/interface.py", line 506, in wrapper_func
return await func(self, *args, **kwargs)
File ".../electrum/electrum/interface.py", line 529, in run
await self.open_session(ssl_context)
File ".../electrum/electrum/interface.py", line 679, in open_session
async with self.taskgroup as group:
File ".../aiorpcX/aiorpcx/curio.py", line 304, in __aexit__
await self.join()
File ".../electrum/electrum/util.py", line 1339, in join
task.result()
File ".../electrum/electrum/synchronizer.py", line 80, in _run_tasks
async with taskgroup as group:
File ".../aiorpcX/aiorpcx/curio.py", line 304, in __aexit__
await self.join()
File ".../electrum/electrum/util.py", line 1339, in join
task.result()
File ".../electrum/electrum/synchronizer.py", line 127, in subscribe_to_address
self.requested_addrs.remove(addr)
KeyError: 'tb1q3wmgf8n5eettnj50pzgnfrrpdpjmwn37x7nzsc5780kk4je9v4hspym8mu'
```
AddressSynchronizer.add_address called synchronizer.add, which would only
schedule adding the addr to the Synchronizer in the next event loop iter.
If during that time, the synchronizer called adb.set_up_to_date(True),
the wallet would falsely believe and advertise itself as up_to_date
(as the wallet would see wallet.synchronize not creating new addresses,
and adb (via synchronizer) telling it is up_to_date).
Moments later, the synchronizer._add_address is finally executed and
up_to_date=False propagates out synchronizer->adb->wallet.
This method is often called when there is already an existing paired
client for the keystore, in which case we can avoid scan_devices() -
which would needlessly take several seconds.
Related to prev commit: multiple (compatible) keystores can now match
with the same HardwareClientBase, so we might as well also allow
reusing existing Clients for the wizard.
- the DeviceMgr no longer uses xpubs to keep track of paired hw devices
- instead, introduce keystore.pairing_code(), based on soft_device_id
- xpubs are now only used in a single place when the actual pairing happens
- motivation is to allow pairing a single device with multiple generic
output script descriptors, not just a single account-level xpub
- as a side-effect, we now allow pairing a device with multiple open
windows simultaneously (if keystores have the same root fingerprint
-- was already the case before if keystores had the same xpub)
Sander van Grieken
ThomasV
SomberNight