ShouldEagerlyRefresh: System.InvalidOperationException: Nullable object must have a value

[HannaHromenko]

FusionCache: EagerExpirationTimestamp Becomes Null After Being Checked

Issue Description

We have been using FusionCache for a while, and from time to time, we encounter the following exception:

System.InvalidOperationException: Nullable object must have a value.  
   at ZiggyCreatures.Caching.Fusion.Internals.FusionCacheInternalUtils.ShouldEagerlyRefresh(IFusionCacheEntry entry)  
   in /_/src/ZiggyCreatures.FusionCache/Internals/FusionCacheInternalUtils.cs:line 124  

From the error message, it is clear that the issue is related to EagerRefresh. Specifically, EagerExpirationTimestamp appears to be null at the time of access, even though there is a check beforehand to verify if it has a value.

Code Analysis

Here’s the FusionCache method where the exception is thrown:

public static bool ShouldEagerlyRefresh(this IFusionCacheEntry? entry)
	{
		if (entry?.Metadata is null)
			return false;

		if (entry.Metadata.EagerExpirationTimestamp.HasValue == false)
			return false;

		if (entry.Metadata.EagerExpirationTimestamp.Value >= DateTimeOffset.UtcNow.UtcTicks) // According to the error log it is thrown here because the value is null even thought we checked it for nullability before
			return false;

		if (entry.IsLogicallyExpired())
			return false;

		return true;
	}

According to the error log, EagerExpirationTimestamp.HasValue initially returns true, but by the time the next line accesses entry.Metadata.EagerExpirationTimestamp.Value, it becomes null, causing the exception.

Observations

We use EagerRefresh in two different places. Both trigger the error:

1. First place where cache is used

private readonly FusionCacheEntryOptions _slidingEntryOptions = new FusionCacheEntryOptions()
    .SetDuration(TimeSpan.FromSeconds(60))
    .SetFailSafe(true)
    .SetFactoryTimeouts(TimeSpan.FromMilliseconds(50))
    .SetEagerRefresh(0.8f);

public async Task<Dictionary<string, Result>> GetCached(string cacheKey, Func<CancellationToken, Task<IEnumerable<Result>>> factory, CancellationToken ct)
{
    return await cache.GetOrSetAsync<Dictionary<string, Result>>(cacheKey,
        async (context, token) =>
        {
            return await factory(ct);
        }, _slidingEntryOptions, token: ct);
}

2. Second place where code is used

private readonly FusionCacheEntryOptions _slidingEntryOptions = new FusionCacheEntryOptions()
    .SetDuration(TimeSpan.FromMinutes(5))
    .SetFailSafe(true)
    .SetFactoryTimeouts(TimeSpan.FromMilliseconds(100))
    .SetEagerRefresh(0.8f);

public async Task<Result?> GetCached(VersionRulesCacheKey cacheKey, Func<string, CancellationToken, Task<Result?>> factory, CancellationToken ct)
{
    return await cache.GetOrSetAsync<VersionRules?>(cacheKey, async (cancellationToken) =>
    {
        return await factory(cacheKey.VersionId, cancellationToken);
    }, _slidingEntryOptions, token: ct);
}

Questions

1. What could cause EagerExpirationTimestamp to become null between the HasValue check and its access?

Any insights would be greatly appreciated! 🚀

[jodydonetti]

Hi @HannaHromenko , sorry for being late, it's been a busy and rough time lately.

FusionCache: EagerExpirationTimestamp Becomes Null After Being Checked

[...]
Any insights would be greatly appreciated! 🚀

Damn it, I think I know exactly why this is happening!
Not so long ago I made a very small optimization regarding eager refresh, whereby once a cache entry enters a fail-safe cycle (after the first fail-safe activation) or even more frequently after an eager refresh kicks in (but before the entry expires), I now reset the eager refresh timestamp. This will reduce the amount of checks and tentative lock acquisition in high usage scenarios. But I did not consider all the edge cases around concurrent usage 🤦‍♂️.

I just created an issue to track this, already fixed it locally and will release it with v2.2.0 coming very soon.

Thanks for spotting it!

[jodydonetti]

I forgot to post it here too: v2.2.0-preview-1 is out 🥳

[jodydonetti]

Hi @HannaHromenko , have you been able to check this out maybe? Thanks!

[HannaHromenko]

Hi @jodydonetti, unfortunately I haven’t checked it yet since we’re trying to avoid deploying the preview versions to production. We’ll wait until v2.2.0 is officially released.

[jodydonetti]

Ok, thanks for the update.

[jodydonetti]

Hi all, v2.2.0 is out 🥳
And for the corresponding LTS-only release, there's v2.0.2.