Etag Implementation #675

hamdaankhalid · 2024-09-18T01:33:25Z

ETag Support

Garnet provides support for ETags on raw strings. By using the ETag-related commands outlined below, you can associate any string-based key-value pair inserted into Garnet with an automatically updated ETag.

Compatibility with non-ETag commands and the behavior of data inserted with ETags are detailed at the end of this document.

SETWITHETAG

Syntax

SETWITHETAG key value [RETAINETAG]

Inserts a key-value string pair into Garnet, associating an ETag that will be updated upon changes to the value.

Options:

RETAINETAG -- Update the Etag associated with the previous key-value pair, while setting the new value for the key. If not etag existed for the previous key this will initialize one.

Response

Integer reply: A response integer indicating the initial ETag value on success.

GETWITHETAG

Syntax

GETWITHETAG key

Retrieves the value and the ETag associated with the given key.

Response

One of the following:

Array reply: An array of two items returned on success. The first item is an integer representing the ETag, and the second is the bulk string value of the key. If called on a key-value pair without ETag, the first item will be nil.
Nil reply: If the key does not exist.

SETIFMATCH

Syntax

SETIFMATCH key value etag

Updates the value of a key if the provided ETag matches the current ETag of the key.

Response

One of the following:

Integer reply: The updated ETag if the value was successfully updated.
Nil reply: If the key does not exist.
Simple string reply: If the provided ETag does not match the current ETag or If the command is called on a record without an ETag a simple string indicating ETag mismatch is returned.

GETIFNOTMATCH

Syntax

GETIFNOTMATCH key etag

Retrieves the value if the ETag associated with the key has changed; otherwise, returns a response indicating no change.

Response

One of the following:

Array reply: If the ETag does not match, an array of two items is returned. The first item is the updated ETag, and the second item is the value associated with the key. If called on a record without an ETag the first item in the array will be nil.
Nil reply: If the key does not exist.
Simple string reply: if the provided ETag matches the current ETag, returns a simple string indicating the value is unchanged.

Compatibility and Behavior with Non-ETag Commands

Below is the expected behavior of ETag-associated key-value pairs when non-ETag commands are used.

MSET, BITOP: These commands will replace an existing ETag-associated key-value pair with a non-ETag key-value pair, effectively removing the ETag.
SET: Only if used with additional option "RETAINETAG" will calling SET update the etag while inserting the new key-value pair over the existing key-value pair.
RENAME: Renaming an ETag-associated key-value pair will reset the ETag to 0 for the renamed key. Unless the key being renamed to already existed before hand, in that case it will retain the etag of the existing key that was the target of the rename.

All other commands will update the etag internally if they modify the underlying data, and any responses from them will not expose the etag to the client. To the users the etag and it's updates remain hidden in non-etag commands.

…tagCmd

TedHartMS

Not finished with review yet but sending a few comments now

libs/common/RespReadUtils.cs

libs/server/Resp/BasicCommands.cs

libs/server/Storage/Functions/MainStore/VarLenInputMethods.cs

hamdaankhalid · 2024-10-02T04:12:14Z

Benchmarking

Etag Impl Branch - Commit `141c870`

BDN

Method	Mean	Error	StdDev	Gen0	Allocated
InlinePing	1.726 us	0.0229 us	0.0191 us	-	-
Set	13.539 us	0.2241 us	0.2096 us	-	-
SetEx	19.572 us	0.2925 us	0.2593 us	-	-
Get	9.680 us	0.1914 us	0.1880 us	-	-
Increment	22.436 us	0.4085 us	0.3821 us	-	-
ZAddRem	112.542 us	1.4627 us	1.2966 us	0.3662	23552 B
LPushPop	102.872 us	0.9256 us	0.8658 us	0.2441	18432 B
SAddRem	96.891 us	0.7661 us	0.7166 us	0.2441	16384 B
HSetDel	132.150 us	2.6330 us	2.1987 us	-	55297 B
MyDictSetGet	177.686 us	1.7051 us	1.5950 us	0.4883	30720 B

RESP

GET

<<<<<<< Benchmark Configuration >>>>>>>>
benchmarkType: Throughput
skipLoad: Disabled
DBsize: 1024
TotalOps: 33554432
Op Benchmark: GET
KeyLength: 1
ValueLength: 8
BatchSize: 4096
RunTime: 15
NumThreads: 1,8,16,32
Auth:
Load Using SET: Disabled
ClientType used for benchmarking: LightClient
TLS: Disabled
ClientHistogram: False
minWorkerThreads: 1000
minCompletionPortThreads: 1000

Generating 8 MSET request batches of size 128 each; total 1024 ops
Request generation complete
maxBytesWritten out of maxBufferSize: 2984/65536
Loading time: 0.002 secs

Operation type: MSET
Num threads: 8
Total time: 84.00ms for 1,024.00 ops
Throughput: 12,190.48 ops/sec

Loaded 1024 keys into the DB >>>

Generating 8192 GET request batches of size 4096 each; total 33554432 ops
Resizing request buffer from 65536 to 131072
Request generation complete
maxBytesWritten out of maxBufferSize: 89851/131072
Loading time: 2.056 secs

Operation type: GET
Num threads: 1
Total time: 15,003.00ms for 110,616,576.00 ops
Throughput: 7,372,963.81 ops/sec

Operation type: GET
Num threads: 8
Total time: 15,010.00ms for 278,532,096.00 ops
Throughput: 18,556,435.44 ops/sec

Operation type: GET
Num threads: 16
Total time: 15,007.00ms for 299,851,776.00 ops
Throughput: 19,980,794.03 ops/sec

Operation type: GET
Num threads: 32
Total time: 15,020.00ms for 291,979,264.00 ops
Throughput: 19,439,365.11 ops/sec

SET

<<<<<<< Benchmark Configuration >>>>>>>>
benchmarkType: Throughput
skipLoad: Disabled
DBsize: 1024
TotalOps: 33554432
Op Benchmark: SET
KeyLength: 1
ValueLength: 8
BatchSize: 4096
RunTime: 15
NumThreads: 1,8,16,32
Auth:
Load Using SET: Disabled
ClientType used for benchmarking: LightClient
TLS: Disabled
ClientHistogram: False
minWorkerThreads: 1000
minCompletionPortThreads: 1000

Generating 8 MSET request batches of size 128 each; total 1024 ops
Request generation complete
maxBytesWritten out of maxBufferSize: 2984/65536
Loading time: 0.003 secs

Operation type: MSET
Num threads: 8
Total time: 18.00ms for 1,024.00 ops
Throughput: 56,888.89 ops/sec

Loaded 1024 keys into the DB >>>

Generating 8192 SET request batches of size 4096 each; total 33554432 ops
Resizing request buffer from 65536 to 131072
Resizing request buffer from 131072 to 262144
Request generation complete
maxBytesWritten out of maxBufferSize: 147195/262144
Loading time: 4.884 secs

Operation type: SET
Num threads: 1
Total time: 15,016.00ms for 112,705,536.00 ops
Throughput: 7,505,696.32 ops/sec

Operation type: SET
Num threads: 8
Total time: 15,003.00ms for 251,764,736.00 ops
Throughput: 16,780,959.54 ops/sec

Operation type: SET
Num threads: 16
Total time: 15,005.00ms for 253,247,488.00 ops
Throughput: 16,877,540.02 ops/sec

Operation type: SET
Num threads: 32
Total time: 15,010.00ms for 237,985,792.00 ops
Throughput: 15,855,149.37 ops/sec

Main Branch - Commit `64636ce`

BDN

Method	Mean	Error	StdDev	Median	Gen0	Allocated
InlinePing	1.766 us	0.0260 us	0.0243 us	1.767 us	-	-
Set	13.370 us	0.1537 us	0.1438 us	13.444 us	-	-
SetEx	19.677 us	0.3127 us	0.2925 us	19.740 us	-	-
Get	10.117 us	0.1089 us	0.1019 us	10.140 us	-	-
Increment	22.060 us	0.3324 us	0.3110 us	22.132 us	-	-
ZAddRem	110.356 us	0.5530 us	0.4902 us	110.402 us	0.3662	23552 B
LPushPop	103.980 us	1.3591 us	1.2713 us	104.172 us	0.2441	18432 B
SAddRem	96.300 us	0.8453 us	0.7494 us	96.195 us	0.2441	16384 B
HSetDel	156.368 us	4.4619 us	12.4380 us	156.181 us	-	55297 B
MyDictSetGet	213.157 us	6.9432 us	20.2536 us	207.258 us	0.4883	30720 B

RESP

GET

<<<<<<< Benchmark Configuration >>>>>>>>
benchmarkType: Throughput
skipLoad: Disabled
DBsize: 1024
TotalOps: 33554432
Op Benchmark: GET
KeyLength: 1
ValueLength: 8
BatchSize: 4096
RunTime: 15
NumThreads: 1,8,16,32
Auth:
Load Using SET: Disabled
ClientType used for benchmarking: LightClient
TLS: Disabled
ClientHistogram: False
minWorkerThreads: 1000
minCompletionPortThreads: 1000

Generating 8 MSET request batches of size 128 each; total 1024 ops
Request generation complete
maxBytesWritten out of maxBufferSize: 2984/65536
Loading time: 0.002 secs

Operation type: MSET
Num threads: 8
Total time: 21.00ms for 1,024.00 ops
Throughput: 48,761.90 ops/sec

Loaded 1024 keys into the DB >>>

Generating 8192 GET request batches of size 4096 each; total 33554432 ops
Resizing request buffer from 65536 to 131072
Request generation complete
maxBytesWritten out of maxBufferSize: 89851/131072
Loading time: 2.322 secs

Operation type: GET
Num threads: 1
Total time: 15,004.00ms for 101,146,624.00 ops
Throughput: 6,741,310.58 ops/sec

Operation type: GET
Num threads: 8
Total time: 15,010.00ms for 276,013,056.00 ops
Throughput: 18,388,611.33 ops/sec

Operation type: GET
Num threads: 16
Total time: 15,005.00ms for 300,077,056.00 ops
Throughput: 19,998,470.91 ops/sec

Operation type: GET
Num threads: 32
Total time: 15,008.00ms for 289,550,336.00 ops
Throughput: 19,293,066.10 ops/sec

SET

<<<<<<< Benchmark Configuration >>>>>>>>
benchmarkType: Throughput
skipLoad: Disabled
DBsize: 1024
TotalOps: 33554432
Op Benchmark: SET
KeyLength: 1
ValueLength: 8
BatchSize: 4096
RunTime: 15
NumThreads: 1,8,16,32
Auth:
Load Using SET: Disabled
ClientType used for benchmarking: LightClient
TLS: Disabled
ClientHistogram: False
minWorkerThreads: 1000
minCompletionPortThreads: 1000

Generating 8 MSET request batches of size 128 each; total 1024 ops
Request generation complete
maxBytesWritten out of maxBufferSize: 2984/65536
Loading time: 0.002 secs

Operation type: MSET
Num threads: 8
Total time: 85.00ms for 1,024.00 ops
Throughput: 12,047.06 ops/sec

Loaded 1024 keys into the DB >>>

Generating 8192 SET request batches of size 4096 each; total 33554432 ops
Resizing request buffer from 65536 to 131072
Resizing request buffer from 131072 to 262144
Request generation complete
maxBytesWritten out of maxBufferSize: 147195/262144
Loading time: 5.168 secs

Operation type: SET
Num threads: 1
Total time: 15,015.00ms for 114,081,792.00 ops
Throughput: 7,597,854.95 ops/sec

Operation type: SET
Num threads: 8
Total time: 15,005.00ms for 247,541,760.00 ops
Throughput: 16,497,284.91 ops/sec

Operation type: SET
Num threads: 16
Total time: 15,021.00ms for 259,411,968.00 ops
Throughput: 17,269,953.27 ops/sec

Operation type: SET
Num threads: 32
Total time: 15,012.00ms for 249,937,920.00 ops
Throughput: 16,649,208.63 ops/sec

TedHartMS

reviewed through MainStoreOps.cs; Tsavorite.core TBD

libs/common/RespReadUtils.cs

libs/server/Resp/BasicCommands.cs

libs/server/Storage/Functions/MainStore/PrivateMethods.cs

libs/server/Storage/Functions/MainStore/RMWMethods.cs

libs/server/Storage/Session/MainStore/MainStoreOps.cs

libs/storage/Tsavorite/cs/src/core/Index/Tsavorite/Implementation/InternalUpsert.cs

TedHartMS · 2024-10-09T16:59:50Z

libs/server/Storage/Session/MainStore/MainStoreOps.cs

+                        Debug.Assert(numItemInArr == 2, "GETWITHETAG output RESP array should be of 2 elements only.");
+
+                        // we know a key-val pair does not have an etag if the first element is not null
+                        // if read nil is successful it will point to the ptrVal otherwise we need to re-read past the etag as RESP int64


These comments don't seem to align with the code; should they be:
// We know a key-val pair has an etag if the first element is not null
// If REadMo; is successful it will increment outputBufCurr to point to the ptrVal; otherwise we need to Read64Into to move outputBufCurr past the etag

TedHartMS · 2024-10-09T17:08:41Z

libs/server/Storage/Session/MainStore/MainStoreOps.cs

+                            bool etagReadingSuccessful = RespReadUtils.Read64Int(out _, ref startOfEtagPtr, end);
+                            Debug.Assert(etagReadingSuccessful, "Etag should have been read succesffuly");
+                            // now startOfEtagPtr is past the etag and points to value 
+                            outputBufCurr = startOfEtagPtr;


Why not simply pass "ref outputBufCurr" on Read64Int? It'd also leave startOfETagPtr where it would be, which would reduce surprise if it's used elsewhere. (It's not, here, and in some cases I use {} to create a temp scope just to have these very-local variables scoped as minimally as possible)

TedHartMS · 2024-10-09T17:13:48Z

libs/server/Storage/Session/MainStore/MainStoreOps.cs

@@ -695,68 +709,75 @@ private unsafe GarnetStatus RENAME(ArgSlice oldKeySlice, ArgSlice newKeySlice, S
                            {


this "hasETag" must always be true now, after your change above to make this the "else" of "!hasETag"

badrishc · 2024-10-17T17:39:48Z

website/docs/commands/etag-commands.md

@@ -0,0 +1,109 @@
+---


We have a section for Garnet specific commands here, perhaps move this documentation there? Or link to it? https://microsoft.github.io/garnet/docs/commands/garnet-specific-commands

badrishc · 2024-10-17T17:40:33Z

website/docs/commands/raw-string.md

@@ -219,6 +219,7 @@ Set **key** to hold the string value. If key already holds a value, it is overwr
 * NX -- Only set the key if it does not already exist.
 * XX -- Only set the key if it already exists.
 * KEEPTTL -- Retain the time to live associated with the key.
+* RETAINETAG -- Update the Etag associated with the previous key-value pair, while setting the new value for the key. If no etag existed on the previous key-value pair this will create the new key-value pair without any etag as well.


call out that this is Garnet-specific, and link to the ETag documentation here.

hamdaankhalid force-pushed the hkhalid/etag-impl branch 4 times, most recently from 77487f1 to d564238 Compare September 25, 2024 07:39

hamdaankhalid marked this pull request as ready for review September 26, 2024 00:35

yrajas requested a review from TedHartMS September 26, 2024 18:28

hamdaankhalidmsft added 16 commits October 1, 2024 13:21

init etag impl

ce7d671

WIP tests

74619d0

complete test coverage and bugfixes

2dbab85

add edge case handling for etag override set

683ead2

Add documentation

7078ec3

fix etag clearing

0e766f5

Format

1f7226d

fix build warnings

c7cb0b5

Update test

3631d20

Update tests for badrish reqs

a079c0f

Update mainstore ops fixing rename bug and remove unnecessary GetForE…

41bdb41

…tagCmd

add read nil test

f6e2fb3

fmt

e89477d

Update docs and add more tests for edgecase

ae5dd43

fix renamenx and etag compat

98e4ec3

Add more etag edge case tests

da48417

hamdaankhalid force-pushed the hkhalid/etag-impl branch from b126e5e to da48417 Compare October 1, 2024 05:30

TedHartMS reviewed Oct 1, 2024

View reviewed changes

libs/common/RespReadUtils.cs Outdated Show resolved Hide resolved

libs/server/Resp/BasicCommands.cs Show resolved Hide resolved

libs/server/Resp/BasicCommands.cs Show resolved Hide resolved

libs/server/Storage/Functions/MainStore/VarLenInputMethods.cs Show resolved Hide resolved

hamdaankhalidmsft added 5 commits October 2, 2024 12:34

remove shifting in network buffer in setifmatch

d581562

Fix named parameters for readability

355bef9

use constants for etag size

8b18e43

update read nil to be a faster

0c340c8

format

141c870

hamdaankhalid requested a review from TedHartMS October 2, 2024 17:13

TedHartMS reviewed Oct 2, 2024

View reviewed changes

TedHartMS requested changes Oct 3, 2024

View reviewed changes

libs/storage/Tsavorite/cs/src/core/Index/Tsavorite/Implementation/InternalUpsert.cs Show resolved Hide resolved

hamdaankhalidmsft added 2 commits October 7, 2024 18:12

pr feedback

a7756c8

Merge branch 'main' into hkhalid/etag-impl

b6ffd7a

hamdaankhalid requested a review from TedHartMS October 8, 2024 01:33

remove redundant ternary

8014242

TedHartMS approved these changes Oct 9, 2024

View reviewed changes

hamdaankhalidmsft added 3 commits October 9, 2024 14:48

update tests for transaction

abc6d8f

Add test to make sure etag data works with recovery scenarios

802ff73

Merge branch 'main' into hkhalid/etag-impl

d11a1e8

badrishc reviewed Oct 17, 2024

View reviewed changes

WIP

8bda79a

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Etag Implementation #675

Etag Implementation #675

hamdaankhalid commented Sep 18, 2024 •

edited

Loading

TedHartMS left a comment

hamdaankhalid commented Oct 2, 2024

TedHartMS left a comment

TedHartMS Oct 9, 2024

TedHartMS Oct 9, 2024

TedHartMS Oct 9, 2024

badrishc Oct 17, 2024

badrishc Oct 17, 2024

		@@ -695,68 +709,75 @@ private unsafe GarnetStatus RENAME(ArgSlice oldKeySlice, ArgSlice newKeySlice, S
		{

Etag Implementation #675

Are you sure you want to change the base?

Etag Implementation #675

Conversation

hamdaankhalid commented Sep 18, 2024 • edited Loading

ETag Support

SETWITHETAG

Syntax

Response

GETWITHETAG

Syntax

Response

SETIFMATCH

Syntax

Response

GETIFNOTMATCH

Syntax

Response

Compatibility and Behavior with Non-ETag Commands

TedHartMS left a comment

Choose a reason for hiding this comment

hamdaankhalid commented Oct 2, 2024

Benchmarking

Etag Impl Branch - Commit 141c870

BDN

RESP

GET

SET

Main Branch - Commit 64636ce

BDN

RESP

GET

SET

TedHartMS left a comment

Choose a reason for hiding this comment

TedHartMS Oct 9, 2024

Choose a reason for hiding this comment

TedHartMS Oct 9, 2024

Choose a reason for hiding this comment

TedHartMS Oct 9, 2024

Choose a reason for hiding this comment

badrishc Oct 17, 2024

Choose a reason for hiding this comment

badrishc Oct 17, 2024

Choose a reason for hiding this comment

hamdaankhalid commented Sep 18, 2024 •

edited

Loading

Etag Impl Branch - Commit `141c870`

Main Branch - Commit `64636ce`