Synapse S3 Storage Maintenance

JIRA: https://sagebionetworks.jira.com/browse/PLFM-510

Related JIRAs:

• https://sagebionetworks.jira.com/browse/PLFM-6165#icft=PLFM-6165

• https://sagebionetworks.jira.com/browse/PLFM-5009

• 
  
  

Related Previous Work:

Introduction

Synapse by default stores most of its data in an encrypted AWS S3 bucket. Additionally external buckets owned by users can be linked to projects and folders so that data can be stored outside the main bucket and its associated costs can be billed separately. Note that synapse supports different types of storage, such a bucket provisioned in google cloud and can link data that is stored elsewhere as long as it’s dereferenceable through a URL. The scope of this document focuses on the S3 storage as it’s the main type of storage currently being adopted.

Whenever a file is uploaded though synapse the reference to the file is maintained in an index, which is a table in an RDS MySQL instance. We refer to a record in this index table as a File Handle. A file handle is simply a pointer to where the physical data is stored and does not provide context where the data is actually used and for this reason there is no explicit access model and instead the user that created the file handle “owns” it (we only allow the owner to delete the file handle or access it directly). Additionally from a user perspective a file handle is immutable and its metadata cannot be updated, this simplifies the internal handling of files in Synapse and provides a generic abstraction over file management.

Once a file handle is created it can be referenced directly in multiple places including file entities, records in a synapse table, user and team profile pictures, wiki attachments, user messages etc. The complete list of type of references is maintained in what we call a file handle associate type:

https://rest-docs.synapse.org/rest/org/sagebionetworks/repo/model/file/FileHandleAssociateType.html.

Through its association a file can be shared with users and teams and downloaded usually through a pre-signed URL.

From a maintenance perspective there are two main categories of data that could lead to potential unwanted costs:

• Unlinked Data
  The current architecture has the advantage that the infrastructure used to manage file uploads is unified in the whole system, but has the drawback that it separates the indexing of the data from the usage of the data, leading to data that can be potentially be un-linked and effectively unused but for which we would still pay the associated storage costs.

• Unindexed Data
  Worth mentioning another category of data in our bucket: data that is instead in the S3 bucket but not indexed in synapse (e.g. a file handle does not exists for a key), this is mostly a concern for the default bucket used by Synapse (as external buckets are managed by other users).

Ideally the system would be designed in a way that the amount of unlinked and unindexed data is kept as low as possible and could self-heal from potential abuse.

Unlinked Data

There are various scenarios when this might happen:

• Updates: for example when a user updates its profile picture or when a version of a file entity is updated with a new file handle. Another example is records in a synapse table that points to a file handle, a table might be updated to use different file handles without deleting the old one.

• Deletions: for example when a project is deleted and after it is purged from the trashcan all the entities in the projects are deleted, the file handles are instead left intact and the data maintained in the bucket.

• No association: a file could be uploaded, a file handle could be created but never linked to anything, making the data inaccessible but to the user itself.

• Data migration: a recent use case might lead to several terabytes of un-linked data, when users want to move their data to an external bucket: users might download from one bucket and re-upload to another bucket (or copying the data over other buckets using the recently developed APIs) updating the file handles in the entities and leaving behind a trail of file handles and data that is not used.

There are various ways that we could tackle this problem:

1. We could introduce an explicit link at the time the data is uploaded and maintain a consistent index enforcing a one to one relationship: this is hard to implement in practice on top of the current architecture, we do not know where the data uploaded will be linked to (we could generate expiring tokens for uploads to be used when linking) and would be a breaking change that would probably take years to be introduced. Additionally we found several millions of file handles already shared between different associations.

2. We could maintain a single index with all the associations and try to keep it consistent: each time a link is established a record is added with the type of association, when an association is broken the index is updated. When the last link is removed the file handle can be flagged as potentially un-linked and archived after a certain amount of time unless it’s linked back. This would require keeping the index up to date (potentially eventually consistent) but it does not work well with deletes as it might be extremely complex to know when an association is broken and communicate the change back to the index: e.g. when a folder is deleted all the files under the folder might be deleted (even though technically we already traverse the hierarchy when purging the trashcan). This brings overhead for each type of association (e.g. the handling needs to be done in each place where file handles are used). The advantage of this approach is that if an association is broken and the link is not removed the worst that can happen is that the data is kept (so no harm). Additionally there are less moving parts and point of failures (compared to the next solution). At the same time we would have the migration problem, this would turn into a big migratable table, but we could technically store it in a separate database that does not migrate, e.g. a potential candidate (there seem to be limitations in how to delete data though) that would scale well would be to store this index in Dynamo DB (using as PK the file handle id and as SK the association).

3. Another solution is instead to periodically scan all the file handle associations and ask which ids are linked, with this information we can build an “index” that can be queried to identify the file handles that are un-linked and archive them for deletion with the possibility to restore them within a given period of time. This approach has the advantage that its implementation has a lower impact on the rest of the system and it is isolated to a specific task potentially reducing the risk of wrong deletions. It’s not immune to mistakes, since when scanning the association a developer could still make a mistake but we can design it so that we can be alerted when mistakes happens before the actual archival/deletion occurs.

In the following we propose a design that revolves around this last solution. There are three main phases that take place:

1. A Discovery phase, where the association are scanned so that links to file handles can be recorded

2. A Detection phase, where with the information from the previous phase we can establish which file handles is not linked

3. An Archival phase, where the file handles that are deemed un-linked are placed into an archive state and will eventually be deleted

File Handle Associations Discovery

The first step is to discover the existing association to file handles. In the backend we maintain already a set of references, generally using dedicated tables with foreign keys back to the file handles table that keep track of the current associations. This goes back to the https://rest-docs.synapse.org/rest/org/sagebionetworks/repo/model/file/FileHandleAssociateType.html and theoretically we keep all the type of associations we have for file handles in order to handle access permissions (e.g. who can download what through which link). This is a bit more complicated in practice, sometimes we do not have a foreign key, but rather a field in a serialized form in the associated record (e.g. profile pictures), or even in a separate database (e.g. File references in synapse tables are maintained as dedicated tables, one for each synapse table).

In the following table we provide the list of known associations to file handles along with a description of how the association is stored:

Note: An interesting observation (Also from the ) is that most of data is referenced by file entities (as expected). Most of our concerns could revolve around un-linking entities rather than other objects. For this it might be actually worth implementing option 2. above instead. We could simply register the links when we encounter them and only take care of un-registering tables and entities links for cleanup.

The idea would be to periodically scan over all the associations and record the last time a file handle was “seen” (e.g. as associated). In this way we can build an index that can be queried to fetch the last time a file handle association was seen so that file handles can be flagged as un-linked.

There are three aspects that require considerations for this process:

1. How to perform the scan

2. Where to store the result

3. When to perform the scan

Scanning

As we can see from the table above the way the associations are stored for each type is not consistent, some types use dedicated tables, other use a column with a field referring to the file handle id and some other store the information embedded in a serialized field. Additionally the size distribution is widely uneven, we have tables with a few thousands associations and tables with millions of records.

The simplest approach would be to simply run a driver job that for each type of association sends an SQS message to starts a sub-job to scan all the associations of that type. Each sub-job performs a scan reading the data in batches. This approach is problematic because while in most of the cases each job might take a few seconds, in some cases it might take several hours (e.g. tables). If a worker goes down while scanning and the job is put back in the queue the job would have to start from scratch.

We instead can split each association in independent sub-jobs that can run in parallel and are only scanning a batch of data (e.g. and dedicate a small fleet of workers for this task), similar to what happens when we migrate (where we scan in batches of 100k records). An idea can be to divide the scan for a given association type in partitions of a given size, for example considering an id used in a table, its min and max value we can divide in sub-jobs of potentially evenly distributed batches. Each batch can be processed by a dedicated worker. This approach can be probably generalized for most of the associations types above providing a default implementation with some variations (e.g. some objects will need to be de-serialized). The tricky part is to divide the scanning in partitions, this can be approximated using in some cases a unique id of the table being scanned, in some other cases (e.g. nodes) we could partition using the file handle id itself.

The idea is to rely on the fault tolerance of both SQS and the worker architecture: if each batch is small enough and a recoverable failure happens the worker can start the same batch assuming that all the information is contained in the SQS message driving the worker.

Synapse tables problem

The most problematic part is scanning the associations in tables:

1. We have several thousand tables, each one with its own file association table

2. Tables are rebuilt every week as are not migratable and in some cases they might actually fail to build all together rendering the file association table useless.

Each table is built using “transactions” which contain the set of changes to the table, we store the history of transactions in a migratable table. If a transaction succeeds we store the change set in S3 as a compressed serialized object. When a table is built it goes through the history of the transactions reading from S3 the change set and applying it to the table. Each table might have several thousands of changes applied to it. Each change set might contain a file handle association and this is when we record the file handle in the table file association table.

One way we could approach scanning the associations for tables is to use the TABLE_ROW_CHANGE table that contains the history of change sets for each table, we can read the changes directly from this table and use it to create sub-jobs that process a batch of changes. Each worker will read from the batch of changes from S3 and look for file handles in the change set itself.

Note that the file handles in this case are freed only when the table is deleted and the changes are dropped. This is unavoidable at the moment due to how the tables are built.

An alternative that would make the scanning much faster is to keep a single migratable table with the file handle associations that is populated when the changes are processed, this table would need to be back-filled with previous data. We can keep this as an optimization if the first approach does not work well.

Storage

We have various options to store this information so that it can be used later in the process.

1. We could store the “linked” status in the file handle table itself along with the timestamp: this would be ideal as we would have to work with only one table and queries would be relatively efficient. Unfortunately this is not possible as the file handle table is our biggest table that is migrated every week and updating the file handles in such a way would most likely lead to unsustainable delays in the migration process. An option would be to move this table in a different “non-migratable” DB, maybe building a dedicated service over file handles, this is a huge task and would required quite a lot of refactoring (e.g. we rely on joying this table).

2. We could store the last time a file handle is linked in a companion migratable table and use it in a similar fashion: this has a similar limitation, it would probably lead to delays in migration as the table would grow substantially over time.

3. We can store this information in another type of external storage.

At the moment option 3. seems the more viable: we can leverage existing technology to store this information as an append only log in S3 directly and perform queries using Athena:

1. We can setup a kinesis firehose stream that delivers the data to a sort of append only log in S3: each time a file handle is scanned we send an event to a kinesis firehose stream, the stream can be setup to target a dedicated S3 bucket and transform the data in a columnar format (e.g. apache parquet) before storage. The records would contain the triple: object type, file handle id, timestamp

2. We can setup a glue table over this S3 destination so that the data can be queried with Athena

3. The bucket can be setup so that old data is automatically removed (e.g. after 60 days)

During the S3 bucket analysis we used Athena to query data in S3 heavily and the results are very promising, joining on mid-size tables directly in S3 (~50M records) is extremely fast (less than 20 seconds) and the results can be stored in S3 directly as target tables. It also has the advantage that once the data is in S3 all the computation can be done externally from a dedicated job (e.g. the data can be joined in S3 to find out the un-liniked file handles and stored as a dedicated table that can be queried separately).

Timing

An important aspect to consider is that the scanning of the associations might take a long time. Even if we do it in parallel we need to limit the amount data scanned at a given time to avoid overloading the DB. We can dedicated a small amount of workers (e.g. 10 workers) to process a batch of data, the data is then sent to a kinesis stream that will eventually deliver it to S3.

This process is not deterministic and failure can occur during the scanning (e.g. a worker goes down, network connectivity is lost etc.). We would still need to have an idea of the scanning progress, or at least when it started and more or less when it completed. For example we do not want to start several scans of the all the associations in parallel, ideally we would scan once in a while (e.g. once a week) and use the start time of the job to filter on the file handles considered for further processing. Before querying the log we would need to know more or less when was the last time a scan finished:

For this we can monitor the stream to check if records are still being processed, if the stream has been empty for more than 24 hours for example we can consider the job that is being running as “finished”. We can also monitor the queue dedicated to the workers, and check if for example errors happens during the process (e.g. messages end up in a dead letter queue). In this case we maintain a table with the scanning jobs and ensure that only one is running at a given time.

Another aspect to consider is how to start a scan:

1. Automatically, according to the status of the current job

2. Externally, using a jenkins job or some other trigger (e.g. a cloud watch event with a lambda) that periodically starts the scan (e.g. every week)

Option 1. would be ideal as the system would be self contained but at the moment Option 2. is preferred to avoid the usual prod vs staging problem.

Un-linked File Handle Detection

Once one or more scans are performed we end up with a log with the last time a file handle was seen as linked. This data will live in S3 and can be queried with Athena. We can now use this information to check if any file handle has not being linked for more than a given amount of time, assuming that a scan was performed recently. Note that we can consider file handles that have been modified before a certain amount of time E.g. we consider only file handles modified more than 30 days ago, this provides a window of for scanning associations multiple times.

Since the file handles are in the main DB we need a way to get this information and we could simply run several queries with Athena with a batch of ids. This is not ideal as it’s a slow process with several limitations, Athena queries are limited in size (e.g. we can probably ask for a 10K batch of ids), we have a limit on the number of concurrent queries that can be performed (e.g. 20/s) and handling the results is tricky (e.g. queries are asynchronous). Where Athena shines is when a single query is performed on a big dataset.

We can instead periodically export the file handle table in S3 and join the data directly using Athena, we can reuse the current infrastructure that replay the change messages on file handles and keep exporting the file handle data in a dedicated glue table using kinesis.

The process is as follow (Note: the data resides in a dedicated bucket that automatically deletes objects after 30 days):

• The file scanner continuously exports associations (currently every 6 days all the associations are scanned, both in prod and staging), using kinesis and a glue table to convert to parquet

• A worker exports the file handle data, including id and updatedOn, using kinesis and a glue table to convert to parquet

• Once a month (1st monday of the month at night) we run an Athena query that joins the two glue tables together to discover all the file handles that do not have an associations (left join with null check), the id of the query is sent to the backend so that results can be processed

• A worker process the athena query results to flag the file handles as unlinked (only handles that are in the available status)

Considered Alternatives:

1. When we run the scanner we update a timestamp of the file handle directly in the DB, a worker will collect the file handles that have not been updated in more than 30 days and flag them as unlinked: This was the original option. It was dismissed because of the risk of trashing migration.

2. Keep a consistent list of links (e.g. add/remove links according to the object that work on file handles): We dismissed it because of the complexity of maintaining a reference count index, additionally we would duplicate data in the DB and migration would suffer from it.

3. Use of S3 infrastructure and objects tags: The idea was to use object tags and a lifecycle configuration, basically trying to tag objects when they were scanned. We dismissed this because updating a tag does not update the update date of the object and lifecycles in s3 buckets are based on it, we would have had to come up with a complex mechanism to update the lifecycle configuration periodically. Additionally the tagging operation costs would add up over time especially if scans are done frequently on all the objects.

See additional notes at

Un-linked File Handle Archival

Deleting user data is a tricky business, especially if we cannot be 100% sure that the data is not used. Instead we propose an approach that goes in stages where first the un-linked data is detected but leaving it accessible and only after a certain amount of time we start archiving it and eventually delete it, with the option to restore it before it is eventually deleted.

We considered various options for archiving the unlinked data:

1. Archive bucket + lifecycle
   The unlinked file handles can be stored in a dedicated bucket starting with the S3 Standard - Infrequent Access storage class. A lifecycle configuration in the archive bucket can be used to move objects after a certain amount of time of their upload to cheaper storage classes and eventually deleted. The process would be as follow:

   • Periodically collect a batch of eligible UNLINKED file handles for which no copy that is still AVAILABLE exist and send a message to a dedicated SQS queue to archive the key

   • A worker uses the transfer manager to copy the object (if bigger than 128KB to avoid paying the overhead in the infrequent access class) to the destination bucket, the file handle status is updated accordingly

   • Through S3 notifications when a file handle is deleted by the lifecycle we can remove the file handle

   Restoring an archived file handle in this case would involve in this case to check the storage class and perform a restore request if in an archive tier, process the S3 notifications when the objects is restored and move the object back to the original location.

2. Intelligent Tiering + tagging
   Instead of using the infrequent access storage class, we can use the new intelligent tiering class (INT). This special class automatically enables monitoring of the objects for access and moves them automatically between standard and infrequent access according to access patterns. If an object is not accessed for 30 days it is moved automatically to INT-infrequent access. If the object is accessed again it is automatically moved back to INT-standard.
   Additionally a new feature allows to enable a specific lifecycle configuration for the INT storage class that moves objects that have not been accessed for more than X days (min 90) to archive tier and after Y days (min 180) to deep archive tier. The archive tiers require a restore operation to access the objects that would move them to the standard class. Additionally this particular lifecycle can be setup to use a rule that uses object tags, so we can enable this only for UNLINKED data.

   The idea would be the following:

   • Move everything in our bucket to INT (we can use a lifecycle configuration for this, using 30 days to avoid moving multipart upload parts of ongoing uploads that are temporary and should not be moved to INT), objects that are smaller than 128KB are not moved (not cost effective)

   • Any object uploaded to our bucket is automatically uploaded as INT (iff > 128KB), this can be part of the storage location metadata

   • Setup a lifecycle configuration for INT so that objects tagged with some special value (e.g. synapse-status=UNLINKED) are moved to archive tier after 90 days and to deep archive tier after 180 days

   • Setup a worker that as in option 1. collects a batch of eligible UNLINKED file handles and tag them appropriately and set the status to ARCHIVED

   • For deletions we can have a worker that process the archived data older than 1 year. Given the recent development of the INT class (e.g. support for archive tier) I expect AWs to implement this for us (e.g. add in the lifecycle the option to delete objects similar to the normal lifecycles), we can wait 1 year before implementing this and see if amazon will deliver.

   Restoring an archived file handle in this case would involve checking the storage class and perform a restore request if in the archive tier and process the S3 notification when the object is restored (no move involved, there is not temporary object for archived data in this case as it is moved automatically to the standard class).
   
   

There are several advantages of option 2. vs option 1: if our objects are all in INT we should see additional cost savings not only for UNLINKED data but also for data that is linked but not accessed while keeping the same durability and performance.

The fact that INT works on the access date rather than creation date is a good fit for our use case where we do not know the access patterns to our data. Additionally the INT class even though it moves object to infrequent storage does not have additional access/retrieval fees for objects while the infrequent access and archive classes such as glacier class can get expensive if we have to access the objects. The biggest advantage is that S3 moves the files around for us according to access date and tagging, we only have to tag objects that we want to move to the archive tier. The restore operation is also simpler (e.g. does not involve copying objects).

The only drawback is that the minimum billable time is 30 days, e.g. if we upload an object in INT and delete it before 30 days we are billed for the full 30 days but I do not think this would be a common scenario we might want to even support. We might want to avoid using INT for the dev account where we create a lot of temporary data and the first time we configure the lifecycle to migrate to INT make sure we set after the rule for 30 days after creation to avoid moving multipart uploads parts that are deleted soon after the upload.

Cost Analysis

In order to determine what is the best option and if it is worth deleting data we collected statistics about the unlinked data and made an estimate of the amount of data that is actually access (hot data) so that we can compare the various options, the following data was collected for the proddata.sagebase.org bucket (To see how refer to ):

• Total Size: 748,386,690,923,161 (680.7 TiB)

• Number of objects: 44,509,789

• Hot Data Count* : 9,802,820

• Hot Data Size*: 314,611,245,904,518 (286.1 TiB)

• Hot Data (>= 128KB) Count: 4,555,052

• Hot Data (>= 128KB) Size: 314,472,063,061,465 (286 TiB)

• Unlinked Data Count: 8,762,805

• Unlinked Data Size: 100,456,586,079,288 (91.36 TiB)

• Unlinked Data Count (>= 128 KB): 2,823,189

• Unlinked Data Size (>= 128 KB): 100,408,552,794,768 (91.32 TiB)

• Monthly Unlinked Count (>= 128KB): 28,808

• Monthly Unlinked Size (>= 128KB): 1,024,577,069,334 (0.93 TiB)

* We considered the file handles downloaded from tables and/or entities from 2020 till now. Does not include other type of downloads, so we will round up to 300TiB of hot data for estimates.

We estimated the monthly cost using the S3 pricing calculator and rounded to 680TiB the total amount of data(T), 300TiB of hot data (H) and 91TiB of unlinked data (U):

1. Leave the data in STD: $15,185

2. Move current unlinked (>=128 KB) to infrequent Access*: $14,394

   1. Using the unlinked data >= 128KB: (680T - 91U) in standard + 91U in IA

   2. Consider an avg of 28,808/month unlinked files for moving data to IA (PUTs cost)

3. Eventually move unlinked data to Glacier deep archive*: $13,323

   1. Using the unlinked data >= 128KB: (680T - 91U) in standard + 91U in Glacier Deep Archive

   2. Consider an avg of 28,808/month unlinked files for moving data to IA (PUTs cost)

   3. Consider the avg of 28,808/month for lifecycle transitions to Glacier and Glacier Deep Archive

4. Move everything to INT**: $11,775

   1. Consider 44% in INT-Standard (300H/680T)

   2. Consider 56% in INT-IA ((680T-300H)/680T)

   3. Includes the 44M objects monitoring fee

5. Move to INT + unlinked eventually in INT-deep archive**: $10,734

   1. Assumes that unlinked data is part of the cold data

   2. Consider 44% in INT-Standard (300H/680T)

   3. Consider 43% in INT-IA ((680TD - 300H - 91U)/680T)

   4. Consider 13% in INT-Deep Archive (91U/680T)

   5. Includes the 44M objects monitoring fee

   6. Assumes that we have 28,808 tags per month (GET/PUT request + tag costs)

6. Move to INT + delete unlinked data***: $10,544

   1. Assumes that unlinked data is part of the cold data

   2. Consider 50% in INT-Standard (300H/(680T - 91))

   3. Consider 50% in INT-IA

   4. Includes the 44M - 8M objects monitoring fee, no tags, no lifecycle transitions

* Note that it does not include the initial cost of moving 2M objects to IA

** Note that it does not include the initial fee for moving all the 44M objects to INT of about $400 (through a lifecycle transition), once an object is in INT there are no fees for other transitions. Additionally in order to detect UNLINKED file handles we use a few additional services heavily, such as kinesis firehose to stream data to S3, Athena for querying, Step Functions to orchestrate and EventBridge for triggering periodically. The service that is most expensive is kinesis firehose as we are pushing a lot of data through, checking the billing for the last few months though this adds up to a few tens of dollars so we excluded it from the estimates, see the cost explorer for kinesis.

*** Note that there is the possibility that a lot of data becomes unlinked in a single month (e.g. a large amount of data is copied in another bucket). We would pay the overhead of this data for a few months that will move to cheaper tiers and eventually archived into deep archive. Given that this is not a common use case (e.g. It might happen once or twice every few years), the complexities around how we store file handles and the current limitations of S3 we do not plan to support a “speed up” of the archival. The user can still request the deletion of their data through the appropriate APIs.

It is clear that just moving everything to INT is cost effective for our use case where our access patterns are unknown since we have data that even if still linked is rarely accessed (e.g. older projects, older versions of entities, older tables, messages etc). On top of it, archiving unlinked data might be worth it even though the cost savings is not comparable to just using the INT storage class, we have around 1TiB of data each month in average that gets unlinked.

Of note is that part of the unlinked data is temporary file handles that synapse creates internally (e.g. table queries, tables csv for uploads etc), the fact that each month we have such a high number (30K) of file handles that are unlinked might be attributed to this. Unfortunately we do not tag this kind of data so we do not know exactly how much of those 1TiB are actually temporary file handles (We do know that at least 20TiB of the current unlinked data is most likely due to this: ).

To solve this we should upload this data either with an special object tag that expires the object through a lifecycle transition or in a dedicated bucket (e.g. another storage location) with a lifecycle that expires everything after 30 days. This temporary objects could be flagged in the database with a special status (e.g. TEMPORARY) so that they are not collected as UNLINKED but rather removed either by S3 notifications (see open questions below) or through a worker.

Another clear point that surfaced is that deleting the unlinked data is not worth the effort, if we move it to the cheapest archive tiering it is actually worth keeping the file handles around just in case we need to recover them.

Proposed Implementation

Archival

In the FILES table we store the status of the file handle which is when created set as AVAILABLE, files that have been detected as UNLINKED are flagged as such. Due to the fact that we allow to create copies of the same file handle that points to the same key we have the following possible scenarios for a file that is flagged as UNLINKED:

1. At least one copy exists for the same key that is still AVAILABLE

2. All of the copies for the same key are flagged as UNLINKED

Note that a file handle can be copied (not the physical data) by different users through a dedicated API, in other words we have multiple file handles pointing to the same bucket and key that could have been “created” by different users.

We want to have a window of time (e.g. 30 days) before we actually archive an object, this provides an additional safety net against potential bugs and misdetections. If a file handle that is not AVAILABLE is accessed the system will send an alert for further investigation, additionally it will not be possible to perform certain operations on a file handle that is not AVAILABLE (e.g. cannot copy it, eventually we will disable any kind of access to not AVAILABLE file handles).

An additional aspect that we need to consider is that the staging database shares the same file handle pointers as production, but at the same time it might be in a state of partial synced truth with prod therefore it is not possible to have an automated scheduled job that is part of the stack to perform the archival.

Instead we propose to add a new dedicated (admin only) asynchronous job that will be invoked by an external process periodically (e.g. a jenkins job):

• The request specifies a batch size and an amount in days to look backwards. This is needed to limit the amount of data that we need to scan per job and avoiding having expensive table scans (Maybe using a separate table for the status might be better, but we would have to refactor every file handle query to use a left join).

• The worker will fetch a batch (LIMIT) of unique keys that:

  • Have been updated more than 30 days ago (let us call this point in time T) and no more than T - <the input days> (if we call this job every 10 minutes we can limit this to a day to avoid big scans)

  • Are flagged as UNLINKED

  • Whose bucket is proddata.sagebase.org (for now we process only our bucket, we can later extend it to other buckets, maybe through a storage location configuration)

• A message is put on a queue that includes a batch of keys (e.g. 100) and T. Another worker will process each key in the message in a separate transaction, for each key:

  • Updates all the file handles for that key older than T and that are UNLINKED so that their status is ARCHIVED

  • Fetches the count of file handles for that key that are AVAILABLE OR (UNLINKED and have been updated after T)

    • If the count is > 0 then the key is still available or another file handle is unlinked but still not past the 30 days window

    • If the count is 0 then we can archive the object in S3: we first get the tag set of the object and then merge it with a new tag such as “synapse-status=archive”. With a configuration for the INT storage class that automatically moves the objects tagged as such in the archive tier and deep archive tier if they were not accessed for more than 90 and 180 days respectively.

  • All the previews of the ARCHIVED file handles can be deleted, as long as they are not used by other available file handles (yeah, this is unfortunately a possibility). The S3 object of the previews is deleted iff the file handle is the last pointing to its key

Note that since we do not care much about keeping track of the job itself the job completes when the batch has finished sending messages to the queue (e.g. not when all the keys in the batch have been processed). If we call this job again too frequently we might risk to put the same key in the queue multiple times but I don’t see any side effects.

Note that we do not delete file handles, in this way we allow them to be restored at a later time if needed by their ids (due to the fact that different users might have created copies of the same file handle). Additionally we want to skip tagging (and therefore moving them to the archive tiers) files smaller than 128KB for two reasons:

• Even if we enable a lifecycle to automatically move objects to INT those will not be moved (we can upload new ones in the INT class though without limitation but they are not automatically move into INT-IA after 30 days)

• The amount of data is irrelevant (a few GB) for being moved to the archive tier tagging it. We would pay more for the get/put requests and object tags than for standard storage (e.g. we have ~6M objects < 128KB, only the tags costs $6/month. Storing the 100GB in Standard is $2.5).

Restore

Since we do not actually delete the data nor the file handles but we potentially store them in an archive tier we need a way to perform a restore (once in the archive tier an object needs an async operation to restore objects that might take a few hours).

We propose to add a new dedicated API that performs the restore through an asynchronous job:

• Takes in input a batch of ids (max 1000) to restore

• For each id checks the file handle status (checking that the user is the creator or admin):

  • If UNLINKED updates it and all the matching file handles to AVAILABLE.

  • If ARCHIVED, fetch the archive status of the S3 object (a head request should be enough, or we can batch this into a list objects):

    • If none then the object is still in INT-IA or INT-Standard, the object can be updated as AVAILABLE along with all the copies and the synapse-status tag can be cleared from the object

    • If present updates it and the matching file handles to RESTORING and sends a restore request to the s3 object, this is an asynchronous call. We enable the S3 notifications and when a restore completes a worker will set the all the file handles that match the keys as AVAILABLE and the synapse-status tag can be cleared from the object

  • Any other status is preserved

Note that even though the job completes some of the file handles might still be RESTORING, for each file handle in input we will include the result in the output of the job with a restore_status: FORBIDDEN, NOT_FOUND, NO_ACTION, RESTORED, RESTORING. We do not wait for the actual restore of the archive tier files, because we would either need to refactor the asynchronous job machinery to support migratable jobs or add dedicated custom code. Hopefully we will never have to use this job.

Note that we need to make sure that when we flag a file handle as UNLINKED we only do so if the status is AVAILABLE and its current updatedOn is < 30 days otherwise we risk to put back the restored objects as UNLINKED.

Open questions

• Why can we make “copies” of file handles in the first place? Maybe we should not allow it in S3 or GC buckets? Who is using this? If we get rid of this does it make it simpler? → This seems to be used heavily by scientists (from DW data), unrealistic to get rid of it and/or deduplicate them.

• I have no idea when the last access date is updated for the INT class nor a way to test it quickly, does tagging the object reset this and moves the object back to INT-Standard? If so we might have a weird lifecycle where an object that is uploaded might go: INT-Standard → (Day 30) INT-IA → UNLINKED → (Day 60, tagging) INT-Standard → (Day 90) INT-IA → (Day 150) INT-Archive → (Day 240) INT-DeepArchive. Basically we might have a month where we go back paying the INT-Standard, in the long run the cost would be amortized since objects will eventually move to the archive tier. → I verified that tagging does NOT in fact push the object back to the INT-Standard. In my own bucket I have objects that are INT-IA, I tagged those objects from the console (therefore fetching the tags as well) and the metrics reported after a day did not change the count of object in INT-IA. This is great news as the lifecycle is now clear and more cost effective: INT-Standard → (Day 30) INT-IA → UNLINKED → (Day 60, Tagging) → (Day 90) INT-Archive → (Day 180) INT-DeepArchive.

• Should we delete the copies of the file handles if at least one is AVAILABLE? If so, if all copies are UNLINKED should we just keep one around? Which one? → No, copies effectively make the ownership a 1 to N relationship

• It is really not clear if lifecycle transitions generate S3 notifications, for example if we wanted to get a notification when a temporary object is automatically expired so that we can remove the file handle record. It looks like this was not possible a few years ago, but the reference in the documentation disappeared, we will have to test this or contact support.

• I didn’t even start thinking about GC or other type of file handles. The vast majority of unlinked data is in prod though.

Unindexed Data

From the it turns out that we have around 7.5TB (this figure didn’t change over time) of data in S3 for which there is no file handle.

This might happen because of various reasons:

• Unfinished multipart uploads: We currently store parts in temporary objects in S3 and the backend later copies the part over to the multipart upload and deletes the temporary parts when the multipart is completed, if the multipart upload is not finished the parts are never deleted. Additionally S3 keeps the multipart upload data “hidden” until the multipart is finished or aborted. As of November 10th we do have 1417823 uncompleted multipart uploads, of which 1414593 were initiated before October 11th (See ). An rough estimate on the current unfinished multipart parts that were uploaded accounts for 2.4TB of data (This number went down to 600GB with the implementation of , we still have 1.5M multipart uploads not finished due to a bug in the code that was later fixed: ).

• Data from staging: Data created in staging is removed after migrations, but of course the data in S3 is left intact.

• Old data already present in the bucket and never cleaned up.

In this case the amount of data compared to the category of un-linked data is not as urgent as it does not seem to grow over time and it’s probably not worth tackling at the moment but still worth mentioning potential solutions. The first point with multi-part uploads is what might be more relevant but we can enable solutions to avoid future costs such as:

• Automatically cleanup unfinished uploads (See )

• Enable a life cycle in the S3 bucket that automatically removes old un-finished multipart uploads (See ).

• Refactor the multipart upload to avoid uploading to temporary objects in the bucket (See )

We enabled the S3 inventory for our bucket and we can write a job that compares the file handle index with the inventory to delete un-indexed data. A simple approach would be to run a script periodically that does the following:

• Queries the inventory with Athena for a certain date range fetching keys in batches (making sure to always have an upper bound of T - 60 days to avoid collecting keys for temporary data such as the multipart upload parts). We can keep track of the date of the object in the last processed batch to be used as the next lower bound.

• For each batch of keys asks the backend (e.g. through a dedicated admin API) which keys do not have file handles

• Saves the response to disk

• For each non matching key deletes the batch of objects (or add a tag so that a lifecycle could expire the objects)

In general we should make sure that any data that ends up in the prod bucket is actually indexed in file handles, for example I would move the temporary objects used for the multipart upload in its own dedicated bucket.