native libc crash when trying to get a realm Instance in multithreaded environment #2207
Comments
|
Hi, can you maybe show how are you calling |
|
Hi @papafe, we are using Xamarin.Forms on Android/IOS, the problem has been reported only on custom devices running Android 5.0/6.0. The relevant piece of code is the following: if (_realmConfiguration == null)
{
lock (Padlock)
{
_realmConfiguration ??= GetRealmConfiguration();
}
}
Realm realm;
try
{
realm = Realm.GetInstance(_realmConfiguration);
}
catch (Realms.Exceptions.RealmException ex)
{
Utils.TraceException(ex,false);
if (!string.IsNullOrEmpty(ex.Message) && ex.Message.Contains("is less than last set version"))
{
lock (Padlock)
{
Realm.DeleteRealm(_realmConfiguration);
_realmConfiguration = GetRealmConfiguration();
}
}
realm = Realm.GetInstance(_realmConfiguration);
}
This is called everytime our app needs a fresh realm instance to work on. The realm configuration is built using the method below: var config = new RealmConfiguration("app_db.db")
{
SchemaVersion = DataModelSchema.MODEL_SCHEMA_VERSION,
ShouldDeleteIfMigrationNeeded = false,
ShouldCompactOnLaunch = (totalBytes, usedBytes) =>
{
try
{
var totalMb = totalBytes / ConvertToMb;
var usedMb = usedBytes / ConvertToMb;
var usedPercentage = (double)usedBytes / totalBytes * 100;
LogBroker.Instance.TraceDebug($"Realm current Total Size: {totalMb:0.##}MB", false);
LogBroker.Instance.TraceDebug($"Realm current Used Size: {usedMb:0.##}MB", false);
LogBroker.Instance.TraceDebug($"Realm current Used Percentage : {usedPercentage:0.##}%", false);
// Compact if the file is over 10MB in size and less than 50% 'used'
const int treshold = 10 * 1024 * 1024;
return totalBytes > treshold && (double)usedBytes / totalBytes < 0.5;
}
catch (Exception e)
{
Utils.TraceException(e);
return false;
}
},
MigrationCallback = (migration, oldSchemaVersion) =>
{
if (oldSchemaVersion < 24)
{
migration.NewRealm.RemoveAll();
return;
}
if (oldSchemaVersion < 26)
{
var newSceneryDatas = Queries.GetAllSceneryData(migration.NewRealm);
for (var i = 0; i < newSceneryDatas.Count(); i++)
{
var newSceneryData = newSceneryDatas.ElementAt(i);
newSceneryData.IsLocked = false;
}
}
if (oldSchemaVersion < 27)
{
var newSystemFunctions = Queries.GetAllSfData(migration.NewRealm);
for (var i = 0; i < newSystemFunctions.Count(); i++)
{
var newSf = newSystemFunctions.ElementAt(i);
newSf.IsActive = 0;
}
}
if (oldSchemaVersion < 28)
{
var newFavouriteModels = Queries.GetAllFavouriteModel(migration.NewRealm);
for (var i = 0; i < newFavouriteModels.Count(); i++)
{
var newFavourite = newFavouriteModels.ElementAt(i);
newFavourite.OrderView = i + 1;
}
}
if (oldSchemaVersion < 30)
{
var bridges = migration.NewRealm.All<Bridge>();
for (var i = 0; i < bridges.Count(); i++)
{
var bridge = bridges.ElementAt(i);
bridge.Connected = false;
}
}
if (oldSchemaVersion < 32)
{
var sfDataVideoentryphones = Queries.GetAllVideoentryphone(migration.NewRealm);
for (var i = 0; i < sfDataVideoentryphones.Count(); i++)
{
var bridge = sfDataVideoentryphones.ElementAt(i);
bridge.IsLocal = false;
}
}
if (oldSchemaVersion < 33)
{
var sfDataVideoentryphones = Queries.GetAllVideoentryphone(migration.NewRealm);
for (var i = 0; i < sfDataVideoentryphones.Count(); i++)
{
var sf = sfDataVideoentryphones.ElementAt(i);
sf.IdAmbientUser = null;
}
}
if (oldSchemaVersion < 35)
{
var newSceneryDatas = Queries.GetAllSceneryData(migration.NewRealm);
var oldSceneryDatas = migration.OldRealm.All("SceneryData");
for (var i = 0; i < newSceneryDatas.Count(); i++)
{
var newSf = newSceneryDatas.ElementAt(i);
var oldSf = oldSceneryDatas.ElementAt(i);
newSf.IdAmbient = (int?)oldSf.IdAmbient;
}
}
if (oldSchemaVersion < 36)
{
var newFavouriteModels = Queries.GetAllFavouriteModel(migration.NewRealm);
for (var i = 0; i < newFavouriteModels.Count(); i++)
{
var newFavourite = newFavouriteModels.ElementAt(i);
newFavourite.IsPlateCamera = false;
}
}
if (oldSchemaVersion < 37)
{
var notificationLedData = migration.NewRealm.All<NotificationLedData>();
foreach (var notification in notificationLedData)
{
notification.IsEnable = false;
}
}
if (oldSchemaVersion < 38)
{
migration.NewRealm.RemoveAll<VideoMessageItemModel>();
}
if (oldSchemaVersion < 39)
{
var newRingtoneSettings = migration.NewRealm.All<RingtonesSettings>().FirstOrDefault();
if(newRingtoneSettings != null)
{
newRingtoneSettings.SettingKey = 0;
}
}
if (oldSchemaVersion < 47)
{
migration.NewRealm.RemoveAll<VideoMessageItemModel>();
migration.NewRealm.RemoveAll<CallHistoryItemModel>();
migration.NewRealm.RemoveAll<TextMessageItemModel>();
}
}
};Thanks for your time! |
|
Can you clarify a little what "custom devices" means in this case? |
|
By custom devices I mean devices with boards manufactured by our client with closed specifications running Android. I don't think this should be an issue for the problem outlined here, but I added this detail for completeness. I should also mention that those devices run on an AOSP with little changes of which I don't know the extension. |
|
If the issue only manifests itself on devices with custom hardware and modifications to the AOSP, there's very little we can do to diagnose or work around it. Since the vendor may have implemented system functionality, such as mutexes, in a non-standard or non-compliant way, we can't offer support for such devices. |
|
I am a colleague of @fede-marsiglia working on this very same issue. I have found that, by looking on the source code of the wrappers library that the only mutex lock used are on the coordinator and on the I think that the case we are seeing here is that we are trying to open a new Realm again after getting the I found that probably it is possible to circumvent this by disabiling realm chache and getting a new realm every time. // If false, always return a new Realm instance, and don't return
// that Realm instance for other requests for a cached Realm. Useful
// for dynamic Realms and for tests that need multiple instances on
// one thread
bool cache = false;Unfortunately the realm C# wrapper does not allow this as the Is it possible that the caching mechanism returns a cached realm that has already been closing, therefore causing the first exception we see and so trying to open a new realm in the same thread causes the second problem due to a bug in the caching mechanism? |
|
There is certainly a possibility for a bug in the caching mechanism, especially if your pattern for opening/disposing Realm instances is unconventional and we didn't do a good job testing for it. Good call going down the reflection approach - note that this may have unintended side effects related to notifications delivery on the main thread, but if it does resolve the original issue, it will definitely narrow down the investigation area. |
|
Thanks for the advice. The code posted by @fede-marsiglia exposes the problem we are facing which in my opinion is more related to the fact that Apart from this the flow we are using is just what it is expected from Realm. Unfortunately the fact that trying to request a new realm in this case fails with the error on the mutext does not let us recover the situation and we cannot even check if the realm is closed. We will try with the workaround but the fact that we are receiving a closed realm is a difficult problem to overcome. Is there some kind of test we can try to reproduce the situation or do some stress test so both us and the Realm team may come to a solution? EDIT: is it possible to keep this issue open while we investigate? |
|
I closed this issue because we can't offer support for custom hardware or modified AOSP distributions because we don't have the means to test those or to ensure that they're implementing basic Unix functionality correctly. It seems that your colleague @lagmac has reported that the issue occurs on a Samsung device and opened #2224 to track it. To avoid splitting the discussion, it's probably best to just use that one. It's hard to suggest a test that will exhibit the problem without knowing how your app works and what are its usage patterns. If you do manage to create a test case that synthetically reproduces the issue somewhat reliably (e.g. >20% of the runs), we'd be happy to take it over from there. |
Goals
Getting a Realm Instance with Realm.GetInstance().
Expected Results
The resulting Realm instance.
Actual Results
Native libc crash, supposedly on mutex use.
Steps to Reproduce
non reproducible, happens in random contexts.
Version of Realm and Tooling
Realm 5.1.1 on Android/Ios
The text was updated successfully, but these errors were encountered: