util: compare raw and encoded keys without explicit decoding #5613

sticnarf · 2019-10-10T03:09:18Z

What have you changed?

This PR adds a function to check if the encoded and raw format refers to the same key.

Motivation

The motivation is that we want to know if a lock is a primary lock. However, inside TiKV we use encoded keys while the stored primary key in the lock is in the raw format.

An alternatives are directly passing raw keys into the transaction layer. But this makes the transaction code uglier. The benchmark result below shows that for a typical key length (30 bytes), it takes just less than 10ns and in most cases, two different keys are different in their tails. So the performance impact is quite small.

What is the type of the changes?

New feature (a change which adds functionality)

How is the PR tested?

Unit test

Does this PR affect documentation (docs) or should it be mentioned in the release notes?

No

Does this PR affect `tidb-ansible`?

No

Refer to a related PR or issue link (optional)

#5575 may utilize it.

Benchmark result if necessary (optional)

test codec::bytes::tests::bench_is_encoded_from       ... bench:         949 ns/iter (+/- 67)
test codec::bytes::tests::bench_is_encoded_from_small ... bench:           5 ns/iter (+/- 0)

Any examples? (optional)


        util: compare raw and encoded keys without explicit decoding

Signed-off-by: Yilin Chen <sticnarf@gmail.com>


        Merge branch 'master' into compare-key


        Merge remote-tracking branch 'upstream/master' into compare-key

BusyJay

Why does index key hardly become primary key? I think it works in the opposite way.

BusyJay · 2019-10-10T12:28:39Z

components/keys/src/types.rs

+        for raw_len in 0..=255 {
+            let raw: Vec<u8> = (0..raw_len).collect();
+            let encoded = Key::from_raw(&raw);
+            assert!(encoded.is_encoded_from(&raw));


Should test negative cases too.

Added. PTAL again

Tests are deliberately repeated in tikv_util::codec and keys, in case we change the implementation in keys to components/codec.

sticnarf · 2019-10-10T14:06:35Z

Why does index key hardly become primary key? I think it works in the opposite way.

My fault...Index keys don't become primary keys in pessimistic transactions. In optimistic transactions, index keys tend to become primary keys. I'll update the main thread.

Anyway, comparing from the end helps return early.


        Add negative cases

Signed-off-by: Yilin Chen <sticnarf@gmail.com>


        Merge branch 'master' into compare-key

youjiali1995 · 2019-10-12T08:09:12Z

components/tikv_util/src/codec/bytes.rs

+                .zip(raw)
+                .all(|(&enc, &raw)| enc == !raw)
+                && encoded[len..encoded.len() - 1].iter().all(|&v| v == 0xff)
+                && encoded[ENC_GROUP_SIZE] == !(ENC_MARKER - pad)


How about comparing it first?

components/tikv_util/src/codec/bytes.rs


        address comments

Signed-off-by: Yilin Chen <sticnarf@gmail.com>

youjiali1995

👍

breeswish · 2019-10-12T11:28:09Z

When will this compare function being used?

sticnarf · 2019-10-12T11:29:58Z

When will this compare function being used?

In #5575, to prevent primary keys of pessimistic transactions from being collapsed.

And maybe in the future, to avoid writing rollback or lock of secondary keys.

breeswish

Looks like this comparing implementation is even slower than simply decoding..

1000 bytes:

test codec::bytes::tests::bench_is_encoded_from       ... bench:         121 ns/iter (+/- 5)
test byte::benches::bench_memcmp_decode_first_asc_large                               ... bench:          92 ns/iter (+/- 4)
test byte::benches::bench_memcmp_decode_first_desc_large                              ... bench:         113 ns/iter (+/- 5)

10000 bytes:

test codec::bytes::tests::bench_is_encoded_from       ... bench:       1,117 ns/iter (+/- 76)
test byte::benches::bench_memcmp_decode_first_asc_large                               ... bench:         749 ns/iter (+/- 45)
test byte::benches::bench_memcmp_decode_first_desc_large                              ... bench:         960 ns/iter (+/- 37)

breeswish · 2019-10-12T13:11:03Z

components/tikv_util/src/codec/bytes.rs

+
+    let mut rev_encoded_chunks = encoded.rchunks_exact(ENC_GROUP_SIZE + 1);
+    // Valid encoded bytes must has complete chunks
+    assert!(rev_encoded_chunks.remainder().is_empty());


I think it would be better to accept invalid bytes but also return false.

Encoded bytes are all generated and used inside TiKV. Therefore, it's either a bug or data corruption if encoded bytes are invalid. So I choose to panic here.

However as a utility function it should not hold that knowledge. You can't prevent this utility function not to be used elsewhere as a common way to compare some user passed-in bytes or TiDB given bytes.

makes sense.

breeswish · 2019-10-12T13:11:03Z

components/tikv_util/src/codec/bytes.rs

+    // Valid encoded bytes must has complete chunks
+    assert!(rev_encoded_chunks.remainder().is_empty());
+
+    // Bytes are compared in reverse order because in TiKV, if two keys are different, the last


Actually this is a TiDB feature instead of TiKV.

Ah, yes...Anyway, I still think it's all right to depend on this feature.

sticnarf · 2019-10-12T13:18:25Z

Looks like this comparing implementation is even slower than simply decoding..

@breeswish If necessary, I think we can do the similar optimizations. But actually this PR helps when we can get the result by check the length or just a few bytes...

breeswish · 2019-10-12T13:32:54Z

@sticnarf If there can be optimizations to do in this PR, then why not do it 🤪I guess the source of slowness is caused from too much abstractions instead of simple and raw array operation..

Note that in TiDB scenario, the key is very short and looks like the effectiveness of this PR can be very trivial.

sticnarf · 2019-10-12T13:39:41Z

@sticnarf If there can be optimizations to do in this PR, then why not do it

My consideration is that this PR is going to be used in transactions to fix a bug in 3.0. Correctness is more important than effectiveness. (And comparing keys is far less costly compared to other operations like rocksdb seek, such optimizations don't improve the final performance much)

Later in master, we can further improve this. There will be more time to test before the next major release.


        Merge remote-tracking branch 'upstream/master' into compare-key


        address comments

Signed-off-by: Yilin Chen <sticnarf@gmail.com>

breeswish · 2019-10-12T13:56:39Z

@sticnarf I don't think bug fix can be an excuse of writing inefficient code, especially considering that it is not that hard to be efficient, as well as this is not a bug that occurs in high priority clients (?).. If there are future plans to improve it, please open an new issue to track it and I can accept it.

breeswish · 2019-10-12T14:02:29Z

components/tikv_util/src/codec/bytes.rs

+
+    let mut rev_encoded_chunks = encoded.rchunks_exact(ENC_GROUP_SIZE + 1);
+    // Valid encoded bytes must has complete chunks
+    assert!(rev_encoded_chunks.remainder().is_empty());


However as a utility function it should not hold that knowledge. You can't prevent this utility function not to be used elsewhere as a common way to compare some user passed-in bytes or TiDB given bytes.

breeswish · 2019-10-12T14:02:29Z

components/tikv_util/src/codec/bytes.rs

+    let raw_chunks = raw.chunks_exact(ENC_GROUP_SIZE);
+    // Check the last chunk first
+    match rev_encoded_chunks.next() {
+        Some(encoded_chunk) if check_single_chunk(encoded_chunk, raw_chunks.remainder()) => {}


what will happen if raw_chunks's length is N*ENC_GROUP_SIZE so that raw_chunks.remainder() is empty?

Then the encoded chunk must be 00 00 00 00 00 00 00 00 FF

breeswish

Except for the panic design, I'm fine with the rest.

breeswish · 2019-10-12T14:26:47Z

components/tikv_util/src/codec/bytes.rs

+                    .iter()
+                    .zip(raw)
+                    .all(|(&enc, &raw)| enc == !raw)
+                && encoded[len..encoded.len() - 1].iter().all(|&v| v == 0xff)


breeswish · 2019-10-12T14:26:47Z

components/tikv_util/src/codec/bytes.rs

+        } else {
+            encoded[ENC_GROUP_SIZE] == (ENC_MARKER - pad)
+                && &encoded[..len] == raw
+                && encoded[len..encoded.len() - 1].iter().all(|&v| v == 0)


!ENC_MARKER


        don't panic when encoded is invalid

Signed-off-by: Yilin Chen <sticnarf@gmail.com>

sticnarf · 2019-10-13T14:57:08Z

@breeswish Panic behavior is changed.

#5641 gives a faster implementation. Could you review that PR and help us evaluate the risk of using that one now?


        make clippy happy

Signed-off-by: Yilin Chen <sticnarf@gmail.com>

breeswish · 2019-10-14T02:25:09Z

components/tikv_util/src/codec/bytes.rs

-/// # Panics
-///
-/// Panics if `encoded` is not valid
+/// Returns whether `encoded` bytes is encoded from `raw`. Returns `false` if `encoded` is invalid.


If encoded is invalid, it must not be encoded from raw. Thus this scenario is already covered by the first sentence.

It's just to clarify that this function can accept invalid encoded bytes :)

sticnarf · 2019-10-14T03:49:02Z

/test

youjiali1995 · 2019-10-14T05:19:09Z

components/keys/src/types.rs

+    ///
+    /// # Panics
+    ///
+    /// Panics if `self` is not a valid encoded key.


Seems it never panic.

Fixed. r? @breeswish


        fix comments for Key::is_encoded_from

Signed-off-by: Yilin Chen <sticnarf@gmail.com>

breeswish · 2019-10-14T06:07:38Z

/merge


        Merge branch 'master' into compare-key

sre-bot · 2019-10-14T06:10:05Z

/run-all-tests

sre-bot · 2019-10-14T06:26:39Z

cherry pick to release-3.1 in PR #5645

sre-bot · 2019-10-14T06:27:49Z

cherry pick to release-3.0 in PR #5646


        util: compare raw and encoded keys without explicit decoding (#5613) (#…

…5646) Signed-off-by: Yilin Chen <sticnarf@gmail.com>


        util: compare raw and encoded keys without explicit decoding (#5613) (#…

…5645) Signed-off-by: Yilin Chen <sticnarf@gmail.com>


        util: compare raw and encoded keys without explicit decoding (tikv#5613)

Signed-off-by: Yilin Chen <sticnarf@gmail.com>

sticnarf added 2 commits Oct 10, 2019

util: compare raw and encoded keys without explicit decoding

Verified

This commit was signed with a verified signature.

sticnarf Yilin Chen

GPG key ID: 353E7ED34BF326E0 Learn about signing commits

Loading status checks…

a72cde0

Signed-off-by: Yilin Chen <sticnarf@gmail.com>

Merge branch 'master' into compare-key

Verified

This commit was created on GitHub.com and signed with a verified signature using GitHub’s key.

GPG key ID: 4AEE18F83AFDEB23 Learn about signing commits

Loading status checks…

3944db5

mapleFU reviewed Oct 10, 2019

View changes

Merge remote-tracking branch 'upstream/master' into compare-key

Loading status checks…

1953617

BusyJay reviewed Oct 10, 2019

View changes

Add negative cases

Verified

This commit was signed with a verified signature.

sticnarf Yilin Chen

GPG key ID: 353E7ED34BF326E0 Learn about signing commits

Loading status checks…

ac24c25

Signed-off-by: Yilin Chen <sticnarf@gmail.com>

sticnarf dismissed mapleFU’s stale review via ac24c25 Oct 10, 2019

Merge branch 'master' into compare-key

Verified

This commit was created on GitHub.com and signed with a verified signature using GitHub’s key.

GPG key ID: 4AEE18F83AFDEB23 Learn about signing commits

Loading status checks…

d8820fe

sticnarf added the C: Util label Oct 10, 2019

youjiali1995 reviewed Oct 12, 2019

View changes

components/tikv_util/src/codec/bytes.rs Outdated Show resolved Hide resolved

address comments

Verified

This commit was signed with a verified signature.

sticnarf Yilin Chen

GPG key ID: 353E7ED34BF326E0 Learn about signing commits

Loading status checks…

fd5784c

Signed-off-by: Yilin Chen <sticnarf@gmail.com>

youjiali1995 reviewed Oct 12, 2019

View changes

sticnarf requested a review from breeswish Oct 12, 2019

breeswish reviewed Oct 12, 2019

View changes

sticnarf added 2 commits Oct 12, 2019

Merge remote-tracking branch 'upstream/master' into compare-key

29dc972

address comments

Verified

This commit was signed with a verified signature.

sticnarf Yilin Chen

GPG key ID: 353E7ED34BF326E0 Learn about signing commits

Loading status checks…

4215ba4

Signed-off-by: Yilin Chen <sticnarf@gmail.com>

sticnarf dismissed youjiali1995’s stale review via 4215ba4 Oct 12, 2019

breeswish reviewed Oct 12, 2019

View changes

don't panic when encoded is invalid

Verified

This commit was signed with a verified signature.

sticnarf Yilin Chen

GPG key ID: 353E7ED34BF326E0 Learn about signing commits

Loading status checks…

5741c61

Signed-off-by: Yilin Chen <sticnarf@gmail.com>

sticnarf mentioned this pull request Oct 13, 2019

util: a faster implementation of is_encoded_from #5641

Open

make clippy happy

Verified

This commit was signed with a verified signature.

sticnarf Yilin Chen

GPG key ID: 353E7ED34BF326E0 Learn about signing commits

Loading status checks…

83d0e43

Signed-off-by: Yilin Chen <sticnarf@gmail.com>

breeswish reviewed Oct 14, 2019

View changes

sticnarf requested a review from youjiali1995 Oct 14, 2019

sticnarf added T: NeedCherryPick-3.0 T: NeedCherryPick-3.1 labels Oct 14, 2019

youjiali1995 reviewed Oct 14, 2019

View changes

fix comments for Key::is_encoded_from

Verified

This commit was signed with a verified signature.

sticnarf Yilin Chen

GPG key ID: 353E7ED34BF326E0 Learn about signing commits

Loading status checks…

04819fb

Signed-off-by: Yilin Chen <sticnarf@gmail.com>

sticnarf dismissed breeswish’s stale review via 04819fb Oct 14, 2019

youjiali1995 approved these changes Oct 14, 2019

View changes

breeswish approved these changes Oct 14, 2019

View changes

sre-bot added the S: CanMerge label Oct 14, 2019

Merge branch 'master' into compare-key

Verified

This commit was created on GitHub.com and signed with a verified signature using GitHub’s key.

GPG key ID: 4AEE18F83AFDEB23 Learn about signing commits

Loading status checks…

a064fde

sre-bot mentioned this pull request Oct 14, 2019

util: compare raw and encoded keys without explicit decoding (#5613) #5645

Merged

sre-bot mentioned this pull request Oct 14, 2019

util: compare raw and encoded keys without explicit decoding (#5613) #5646

Merged

sticnarf added a commit to sticnarf/tikv that referenced this pull request Oct 27, 2019

util: compare raw and encoded keys without explicit decoding (tikv#5613)

61ff13f

Signed-off-by: Yilin Chen <sticnarf@gmail.com>

tikv/tikv

Join GitHub today

util: compare raw and encoded keys without explicit decoding #5613

util: compare raw and encoded keys without explicit decoding #5613

Conversation

sticnarf commented Oct 10, 2019 • edited

What have you changed?

Motivation

What is the type of the changes?

How is the PR tested?

Does this PR affect documentation (docs) or should it be mentioned in the release notes?

Does this PR affect tidb-ansible?

Refer to a related PR or issue link (optional)

Benchmark result if necessary (optional)

Any examples? (optional)

BusyJay left a comment

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

sticnarf commented Oct 10, 2019

This comment has been minimized.

youjiali1995 left a comment

This comment has been minimized.

breeswish commented Oct 12, 2019

This comment has been minimized.

sticnarf commented Oct 12, 2019 • edited

breeswish left a comment

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

sticnarf commented Oct 12, 2019 • edited

This comment has been minimized.

breeswish commented Oct 12, 2019

This comment has been minimized.

sticnarf commented Oct 12, 2019

This comment has been minimized.

breeswish commented Oct 12, 2019 • edited

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

breeswish left a comment

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

sticnarf commented Oct 13, 2019

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

sticnarf commented Oct 14, 2019

This comment has been minimized.

This comment has been minimized.

sticnarf Oct 14, 2019 • edited

This comment has been minimized.

breeswish commented Oct 14, 2019

This comment has been minimized.

sre-bot commented Oct 14, 2019

This comment has been minimized.

sre-bot commented Oct 14, 2019

This comment has been minimized.

sre-bot commented Oct 14, 2019

sticnarf commented Oct 10, 2019 •

edited

Does this PR affect `tidb-ansible`?

sticnarf commented Oct 12, 2019 •

edited

sticnarf commented Oct 12, 2019 •

edited

breeswish commented Oct 12, 2019 •

edited

sticnarf Oct 14, 2019 •

edited