Regression: Garnet 1.0.49 fails on unknown commands in Lua

[nightroman]

Describe the bug

Garnet 1.0.49 has issues with processing unknown commands in Lua (1.0.48 works more as expected).

Steps to reproduce the bug

With Garnet 1.0.48 using StackExchange.Redis ScriptEvaluate, we get:

• invoking redis.call("missing") returns an empty string
• invoking return redis.call("missing") returns the string ERR unknown command

With Garnet 1.0.49 the above calls result in exceptions and the server log shows the following records (they do not look as something by design):

05::14::27 fail: Session[0] [127.0.0.1:61856] [00480155] During Lua script execution System.Runtime.InteropServices.SEHException (0x80004005): External component has thrown an exception.    at KeraLua.NativeMethods.lua_error(IntPtr luaState)    at Garnet.server.LuaStateWrapper.RaiseErrorFromStack() in /_/libs/server/Lua/LuaStateWrapper.cs:line 476    at Garnet.server.LuaRunner.LuaStaticError(Int32 constStringRegistryIndex) in /_/libs/server/Lua/LuaRunner.cs:line 656    at Garnet.server.LuaRunner.ProcessResponse(Byte* ptr, Int32 length) in /_/libs/server/Lua/LuaRunner.cs:line 698    at Garnet.server.LuaRunner.ProcessCommandFromScripting[TGarnetApi](TGarnetApi api) in /_/libs/server/Lua/LuaRunner.cs:line 634

05::14::27 fail: Session[0] [127.0.0.1:61856] [00480155] Error executing Lua script System.Runtime.InteropServices.SEHException (0x80004005): External component has thrown an exception.    at KeraLua.NativeMethods.lua_error(IntPtr luaState)    at Garnet.server.LuaStateWrapper.RaiseErrorFromStack() in /_/libs/server/Lua/LuaStateWrapper.cs:line 476    at Garnet.server.LuaStateWrapper.RaiseError(String msg) in /_/libs/server/Lua/LuaStateWrapper.cs:line 466    at Garnet.server.LuaRunner.ProcessCommandFromScripting[TGarnetApi](TGarnetApi api) in /_/libs/server/Lua/LuaRunner.cs:line 642    at Garnet.server.LuaRunner.garnet_call_txn(IntPtr luaStatePtr) in /_/libs/server/Lua/LuaRunner.cs:line 502    at KeraLua.NativeMethods.lua_pcallk(IntPtr luaState, Int32 nargs, Int32 nresults, Int32 errorfunc, IntPtr ctx, IntPtr k)    at KeraLua.Lua.PCall(Int32 arguments, Int32 results, Int32 errorFunctionIndex)    at Garnet.server.LuaRunner.RunCommon[TResponse](TResponse& resp) in /_/libs/server/Lua/LuaRunner.cs:line 1084    at Garnet.server.LuaRunner.RunForSession(Int32 count, RespServerSession outerSession) in /_/libs/server/Lua/LuaRunner.cs:line 853    at Garnet.server.RespServerSession.ExecuteScript(Int32 count, LuaRunner scriptRunner) in /_/libs/server/Lua/LuaCommands.cs:line 264

Expected behavior

The above Lua scripts are intentionally not quite right, using unknown commands.
They should probably cause some errors, e.g. like in Garnet 1.0.48 but not like in 1.0.49

Release version

1.0.49

IDE

.NET 9

OS version

Windows 11

[kevin-montrose]

Hmm, I can't quite reproduce what you're seeing in .49 (commit ce21c248f084744e45bbff08d0ecce0a51326cca). Specifically the log lines.

As to the responses, I think those are correct in .49 and were wrong in earlier Garnet. Comparing to Redis, ScriptEvaluate("redis.call(\"missing\")") returns an ERR the same way ScriptEvaluate("return redis.call(\"missing\")") does.

Could you share a repro script that produces those log lines?

[nightroman]

@kevin-montrose If I use GarnetServer from the latest repo with --lua --lua-transaction-mode then I cannot repro the issue. And the errors are "ERR Unknown Redis command called from script" for both cases, as you say.

I observe problems in a server built from the NuGet package. Have you tried this scenario, by any chance?

[nightroman]

My server built from the NuGet package is this https://github.com/nightroman/FarNet.Redis/tree/main/src/GarnetServer
with

    <PackageReference Include="Microsoft.Garnet" Version="1.0.49" />

[badrishc]

Looks liket the nuget has not packaged up the lua binaries. The nuget spec is in the csproj now: https://github.com/microsoft/garnet/blob/main/libs/host/Garnet.host.csproj

Also we need to check whether dotnet-tool works: https://github.com/microsoft/garnet/blob/main/main/GarnetServer/GarnetServer.csproj

[nightroman]

FWIW My normal Lua tests pass (not many, a couple). It just abnormal Lua scripts calling missing Redis commands misbehave.

https://github.com/nightroman/FarNet.Redis/blob/main/tests/Scripting.test.ps1
(the last is commented out due to 1.0.49 issues)

[kevin-montrose]

@kevin-montrose If I use GarnetServer from the latest repo with --lua --lua-transaction-mode then I cannot repro the issue. And the errors are "ERR Unknown Redis command called from script" for both cases, as you say.

I observe problems in a server built from the NuGet package. Have you tried this scenario, by any chance?

I tested with the repo at the .49 commit, I'll try with the nuget package.

[kevin-montrose]

Brand new project hosting .49 (command line args (--lua --lua-transaction-mode --logger-level Trace), I cannot repro the exceptions:

Using SE.Redis 2.8.22 in that LinqPad script, .NET 8 for the console app, tested Debug and Release configs.