If one request failed, the whole batch would fail.
This caused issue as one of the contract is constantly returning an
error now, and essentially there was not way to fetch balance.
Also extend the timeout to 20s as we throw 165 request to Infura in one
go and it takes its time to reply to those, although it seems like we
should batch them on our side instead of sending them all cuncurrently.
StartWallet was called before service initialization.
After the recent changes this call was moved after initialization, but
the geth system automatically start services.
This meant that `IsStarted()` returned true, although the reactor was
not started, and only after calling `StopWallet()` and `StartWallet()`
again the system would reach the right state.
This commit changes the behavior so that we only check whether the
reactor has been started when calling `IsStarted()` and we allow
multiple calls to `Start()` on the signal service, which won't return an
error (it's a noop if callled multiple times).
- unused API methods are removed
- some unusued code is removed too
- API docs are updated
That's just a portion of clean up that should be done,
but the rest of it will probably happen in different PR
with changes to the way how we watch to chain updates.
- In order to avoid handling of the reorganized blocks we use an offset
from the latest known block when start listening to new blocks. Before
this commit the offset was 15 blocks for all networks. This offset is
too big for mainnet and causes noticeable delay of marking a transfer as
confirmed in Status (comparing to etherscan). So it was changed to be 5
blocks on mainnet and is still 15 blocks on other networks.
- Also before this commit all new blocks were handled one by one with
network specific interval (10s for mainnet), which means that in case of
lost internet connection or application suspension (happens on iOS)
receiving of new blocks would be paused and then resumed with the same
"speed" - 1 blocks per 10s. In case if that pause is big enough the
application would never catch up with the latest block in the network,
and this also causes the state of transfers to be delayed in the
application. In this commit in case if there was more than 40s delay
after receiving of the previous block the whole history in range between
the previous received block and ("latest"-reorgeSafetyDepth) block is
checked at once and app catches up with a recent state of the chain.
*** How it worked before this PR on multiaccount creation:
- On multiacc creation we scanned chain for eth and erc20 transfers. For
each address of a new empty multiaccount this scan required
1. two `eth_getBalance` requests to find out that there is no any
balance change between zero and the last block, for eth transfers
2. and `chain-size/100000` (currently ~100) `eth_getLogs` requests,
for erc20 transfers
- For some reason we scanned an address of the chat account as well, and
also accounts were not deduplicated. So even for an empty multiacc we
scanned chain twice for each chat and main wallet addresses, in result
app had to execute about 400 requests.
- As mentioned above, `eth_getBalance` requests were used to check if
there were any eth transfers, and that caused empty history in case
if user already used all available eth (so that both zero and latest
blocks show 0 eth for an address). There might have been transactions
but we wouldn't fetch/show them.
- There was no upper limit for the number of rpc requests during the
scan, so it could require indefinite number of requests; the scanning
algorithm was written so that we persisted the whole history of
transactions or tried to scan form the beginning again in case of
failure, giving up only after 10 minutes of failures. In result
addresses with sufficient number of transactions would never be fully
scanned and during these 10 minutes app could use gigabytes of
internet data.
- Failures were caused by `eth_getBlockByNumber`/`eth_getBlockByHash`
requests. These requests return significantly bigger responses than
`eth_getBalance`/`eth_transactionsCount` and it is likely that
execution of thousands of them in parallel caused failures for
accounts with hundreds of transactions. Even for an account with 12k
we could successfully determine blocks with transaction in a few
minutes using `eth_getBalance` requests, but `eth_getBlock...`
couldn't be processed for this acc.
- There was no caching for for `eth_getBalance` requests, and this
caused in average 3-4 times more such requests than is needed.
*** How it works now on multiaccount creation:
- On multiacc creation we scan chain for last ~30 eth transactions and
then check erc20 in the range where these eth transactions were found.
For an empty address in multiacc this means:
1. two `eth_getBalance` transactions to determine that there was no
balance change between zero and the last block; two
`eth_transactionsCount` requests to determine there are no outgoing
transactions for this address; total 4 requests for eth transfers
2. 20 `eth_getLogs` for erc20 transfers. This number can be lowered,
but that's not a big deal
- Deduplication of addresses is added and also we don't scan chat
account, so a new multiacc requires ~25 (we also request latest block
number and probably execute a few other calls) request to determine
that multiacc is empty (comparing to ~400 before)
- In case if address contains transactions we:
1. determine the range which contains 20-25 outgoing eth/erc20
transactions. This usually requires up to 10 `eth_transactionCount`
requests
2. then we scan chain for eth transfers using `eth_getBalance` and
`eth_transactionCount` (for double checking zero balances)
3. we make sure that we do not scan db for more than 30 blocks with
transfers. That's important for accounts with mostly incoming
transactions, because the range found on the first step might
contain any number of incoming transfers, but only 20-25 outgoing
transactions
4. when we found ~30 blocks in a given range, we update initial
range `from` block using the oldest found block
5. and now we scan db for erc20transfers using `eth_getLogs`
`oldest-found-eth-block`-`latest-block`, we make not more than 20 calls
6. when all blocks which contain incoming/outgoing transfers for a
given address are found, we save these blocks to db and mark that
transfers from these blocks are still to be fetched
7. Then we select latest ~30 (the number can be adjusted) blocks from
these which were found and fetch transfers, this requires 3-4
requests per transfer.
8. we persist scanned range so that we know were to start next time
9. we dispatch an event which tells client that transactions are found
10. client fetches latest 20 transfers
- when user presses "fetch more" button we check if app's db contains next
20 transfers, if not we scan chain again and return transfers after
small fixes
Wallet database refactored so that every query ensures isolation by the network id.
Network id provided when database object is created, thus it is transparent to other parts
of the wallet module.
Additionally every uniqueness index is changed to ensure that it doesn't prevent adding
object with same id but from a different network.
* Store tx and receipt in db and cast it to TransferView on read
* Store Log instead of log index
* Use contract from log and bring back address field
* Add tx status and id fields
[services/wallet] Several changes in API after feedback
- Timestamp from block header stored in blocks table and added to each transfers
- From field is computed from signature to offload this computation from client side
- `history` event is added back, so that client can ignore historical blocks when watching
only for new blocks
- block number and timestamp are marshalled in hex. consistent with ethereum data structures
Gheorghe Pinzaru
Andrea Maria Piana
andrey
RichΛrd
Roman Volosovskyi
Adam Babik
Pedro Pombeiro
André Medeiros
dmitry