-
Notifications
You must be signed in to change notification settings - Fork 1.7k
Problem: AuRa's unsafeties around step duration #7282
Conversation
It looks like @yrashk hasn'signed our Contributor License Agreement, yet.
You can read and sign our full Contributor License Agreement at the following URL: https://cla.parity.io Once you've signed, plesae reply to this thread with Many thanks, Parity Technologies CLA Bot |
The test failure seems to be unrelated to the patch:
|
[clabot:check] |
It looks like @yrashk signed our Contributor License Agreement. 👍 Many thanks, Parity Technologies CLA Bot |
@@ -1073,4 +1109,50 @@ mod tests { | |||
assert_eq!(aura.maximum_uncle_count(1), 0); | |||
assert_eq!(aura.maximum_uncle_count(100), 0); | |||
} | |||
|
|||
#[test] |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
indentation issue here
LGTM except for indentation in one place. |
edc57f2
to
a38465b
Compare
@rphmeier Robert, I've addressed your comments. Please let me know if you have any other thoughts. |
Firstly, `Step.duration_remaining` casts it to u32, unnecesarily limiting it to 2^32. While theoretically this is "good enough" (at 3 seconds steps it provides room for a little over 400 years), it is still a lossy way to calculate the remaining time until the next step. Secondly, step duration might be zero, triggering division by zero in `Step.calibrate` Solution: rework the code around the fact that duration is typically in single digits and never grows, hence, it can be represented by a much narrower range (u16) and this highlights the fact that multiplying u64 by u16 will only result in an overflow in even further future, at which point we should panic informatively (if anybody's still around) Similarly, panic when it is detected that incrementing the step counter wrapped around on the overflow of usize. As for the division by zero, prevent it by making zero an invalid value for step duration. This will make AuRa log the constraint mismatch and panic (after all, what purpose would zero step duration serve? it makes no sense within the definition of the protocol, as finality can only be achieved as per the specification if messages are received within the step duration, which would violate the speed of light and other physical laws in this case).
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Tiny grumble.
step_end - now | ||
} else { | ||
Duration::from_secs(0) | ||
let expected_seconds = (self.load() as u64) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
let expected_seconds = (self.load() as u64)
.checked_add(1)
.and_then(|ctr| ctr.checked_mul(self.duration as u64))
.map(Duration::from_secs);
match expected_seconds {
Some(step_end) if step_end > now => {
step_end - now
},
Some(_) => {
Duration::from_secs(0)
},
None => {
...
}
}
Firstly, `Step.duration_remaining` casts it to u32, unnecesarily limiting it to 2^32. While theoretically this is "good enough" (at 3 seconds steps it provides room for a little over 400 years), it is still a lossy way to calculate the remaining time until the next step. Secondly, step duration might be zero, triggering division by zero in `Step.calibrate` Solution: rework the code around the fact that duration is typically in single digits and never grows, hence, it can be represented by a much narrower range (u16) and this highlights the fact that multiplying u64 by u16 will only result in an overflow in even further future, at which point we should panic informatively (if anybody's still around) Similarly, panic when it is detected that incrementing the step counter wrapped around on the overflow of usize. As for the division by zero, prevent it by making zero an invalid value for step duration. This will make AuRa log the constraint mismatch and panic (after all, what purpose would zero step duration serve? it makes no sense within the definition of the protocol, as finality can only be achieved as per the specification if messages are received within the step duration, which would violate the speed of light and other physical laws in this case).
Firstly, `Step.duration_remaining` casts it to u32, unnecesarily limiting it to 2^32. While theoretically this is "good enough" (at 3 seconds steps it provides room for a little over 400 years), it is still a lossy way to calculate the remaining time until the next step. Secondly, step duration might be zero, triggering division by zero in `Step.calibrate` Solution: rework the code around the fact that duration is typically in single digits and never grows, hence, it can be represented by a much narrower range (u16) and this highlights the fact that multiplying u64 by u16 will only result in an overflow in even further future, at which point we should panic informatively (if anybody's still around) Similarly, panic when it is detected that incrementing the step counter wrapped around on the overflow of usize. As for the division by zero, prevent it by making zero an invalid value for step duration. This will make AuRa log the constraint mismatch and panic (after all, what purpose would zero step duration serve? it makes no sense within the definition of the protocol, as finality can only be achieved as per the specification if messages are received within the step duration, which would violate the speed of light and other physical laws in this case).
Firstly, `Step.duration_remaining` casts it to u32, unnecesarily limiting it to 2^32. While theoretically this is "good enough" (at 3 seconds steps it provides room for a little over 400 years), it is still a lossy way to calculate the remaining time until the next step. Secondly, step duration might be zero, triggering division by zero in `Step.calibrate` Solution: rework the code around the fact that duration is typically in single digits and never grows, hence, it can be represented by a much narrower range (u16) and this highlights the fact that multiplying u64 by u16 will only result in an overflow in even further future, at which point we should panic informatively (if anybody's still around) Similarly, panic when it is detected that incrementing the step counter wrapped around on the overflow of usize. As for the division by zero, prevent it by making zero an invalid value for step duration. This will make AuRa log the constraint mismatch and panic (after all, what purpose would zero step duration serve? it makes no sense within the definition of the protocol, as finality can only be achieved as per the specification if messages are received within the step duration, which would violate the speed of light and other physical laws in this case).
Firstly, `Step.duration_remaining` casts it to u32, unnecesarily limiting it to 2^32. While theoretically this is "good enough" (at 3 seconds steps it provides room for a little over 400 years), it is still a lossy way to calculate the remaining time until the next step. Secondly, step duration might be zero, triggering division by zero in `Step.calibrate` Solution: rework the code around the fact that duration is typically in single digits and never grows, hence, it can be represented by a much narrower range (u16) and this highlights the fact that multiplying u64 by u16 will only result in an overflow in even further future, at which point we should panic informatively (if anybody's still around) Similarly, panic when it is detected that incrementing the step counter wrapped around on the overflow of usize. As for the division by zero, prevent it by making zero an invalid value for step duration. This will make AuRa log the constraint mismatch and panic (after all, what purpose would zero step duration serve? it makes no sense within the definition of the protocol, as finality can only be achieved as per the specification if messages are received within the step duration, which would violate the speed of light and other physical laws in this case).
* Merge pull request #7368 from paritytech/td-future-blocks Wait for future blocks in AuRa * Advance AuRa step as far as we can and prevent invalid blocks. (#7451) * Advance AuRa step as far as we can. * Wait for future blocks. * Problem: AuRa's unsafeties around step duration (#7282) Firstly, `Step.duration_remaining` casts it to u32, unnecesarily limiting it to 2^32. While theoretically this is "good enough" (at 3 seconds steps it provides room for a little over 400 years), it is still a lossy way to calculate the remaining time until the next step. Secondly, step duration might be zero, triggering division by zero in `Step.calibrate` Solution: rework the code around the fact that duration is typically in single digits and never grows, hence, it can be represented by a much narrower range (u16) and this highlights the fact that multiplying u64 by u16 will only result in an overflow in even further future, at which point we should panic informatively (if anybody's still around) Similarly, panic when it is detected that incrementing the step counter wrapped around on the overflow of usize. As for the division by zero, prevent it by making zero an invalid value for step duration. This will make AuRa log the constraint mismatch and panic (after all, what purpose would zero step duration serve? it makes no sense within the definition of the protocol, as finality can only be achieved as per the specification if messages are received within the step duration, which would violate the speed of light and other physical laws in this case). * Fix tests. * detect different node, same-key signing in aura (#7245) * detect different node, same-key signing in aura * reduce scope of warning
* Merge pull request #7368 from paritytech/td-future-blocks Wait for future blocks in AuRa * Fix tracing failed calls. * Problem: sending any Whisper message fails The error is "PoW too low to compete with other messages" This has been previously reported in #7144 Solution: prevent the move semantics The source of the error is in PoolHandle.relay implementation for NetPoolHandle. Because of the move semantics, `res` variable is in fact copied (as it implements Copy) into the closure and for that reason, the returned result is always `false. * Merge pull request #7433 from paritytech/td-strict-config Strict config parsing * Problem: AuRa's unsafeties around step duration (#7282) Firstly, `Step.duration_remaining` casts it to u32, unnecesarily limiting it to 2^32. While theoretically this is "good enough" (at 3 seconds steps it provides room for a little over 400 years), it is still a lossy way to calculate the remaining time until the next step. Secondly, step duration might be zero, triggering division by zero in `Step.calibrate` Solution: rework the code around the fact that duration is typically in single digits and never grows, hence, it can be represented by a much narrower range (u16) and this highlights the fact that multiplying u64 by u16 will only result in an overflow in even further future, at which point we should panic informatively (if anybody's still around) Similarly, panic when it is detected that incrementing the step counter wrapped around on the overflow of usize. As for the division by zero, prevent it by making zero an invalid value for step duration. This will make AuRa log the constraint mismatch and panic (after all, what purpose would zero step duration serve? it makes no sense within the definition of the protocol, as finality can only be achieved as per the specification if messages are received within the step duration, which would violate the speed of light and other physical laws in this case). * Merge pull request #7437 from paritytech/a5-chains-expanse Remove expanse chain * Expanse Byzantium update w/ correct metropolis difficulty increment divisor (#7463) * Byzantium Update for Expanse Here the changes go. Hope I didnt miss anything. * expip2 changes - update duration limit * Fix missing EXPIP-2 fields * Format numbers as hex * Fix compilation errors * Group expanse chain spec fields together * Set metropolisDifficultyIncrementDivisor for Expanse * Revert #7437 * Add Expanse block 900_000 hash checkpoint * Advance AuRa step as far as we can and prevent invalid blocks. (#7451) * Advance AuRa step as far as we can. * Wait for future blocks. * fixed panic when io is not available for export block, closes #7486 (#7495) * Update Parity Mainnet Bootnodes (#7476) * Update Parity Mainnet Bootnodes * Replace the Azure HDD bootnodes with the new ones :) * Use https connection (#7503) Use https when connecting to etherscan.io API for price-info * Expose default gas price percentile configuration in CLI (#7497) * Expose gas price percentile. * Fix light eth_call. * fix gas_price in light client
Firstly,
Step.duration_remaining
casts it to u32, unnecesarilylimiting it to 2^32. While theoretically this is "good enough" (at 3
seconds steps it provides room for a little over 400 years), it is
still a lossy way to calculate the remaining time until the next step.
Secondly, step duration might be zero, triggering division by zero
in
Step.calibrate
Solution: rework the code around the fact that duration is
typically in single digits and never grows, hence, it can be represented
by a much narrower range (u8) and this highlights the fact that
multiplying u64 by u8 will only result in an overflow in even further
future, at which point we should panic informatively (if anybody's
still around)
Similarly, panic when it is detected that incrementing the step
counter wrapped around on the overflow of usize.
As for the division by zero, prevent it by making zero an invalid
value for step duration. This will make AuRa log the constraint
mismatch and panic (after all, what purpose would zero step duration
serve? it makes no sense within the definition of the protocol,
as finality can only be achieved as per the specification
if messages are received within the step duration, which would violate
the speed of light and other physical laws in this case).