support multiple storage backends, including amazon s3

[zooko]

The focus of this ticket is (now) adapting the existing codebase to use multiple backends, rather than supporting any particular backend. We already have one backend -- the filesystem backend -- which I think should be a plugin in the same sense that the others will be plugins (i.e.: other code in tahoe-lafs can interact with a filesystem plugin without caring very much about how or where it is storing its files -- otherwise it doesn't seem very extensible). If you accept this, then we'd need to figure out what a backend plugin should look like. There is backend-independent logic in the current server implementation that we wouldn't want to duplicate in every other backend implementation. To address this, we could start by refactoring the existing code that reads or writes shares on disk, to use a local backend implementation supporting an IStorageProvider interface (probably a fairly simplistic filesystem-ish API). (This involves changing the code in src/allmydata/storage/server.py​ that reads from local disk in its _iter_share_files() method, and also changing storage/shares.py, storage/immutable.py​, and storage/mutable.py that write shares to local disk.) At this point all the existing tests should still pass, since we haven't actually changed the behaviour. Then we have to add the ability to configure new storage providers. This involves figuring out how to map user configuration choices to what actually happens when a node is started, and how the credentials needed to log into a particular storage backend should be specified. The skeletal RIStorageServer would instantiate its IStorageProvider based on what the user configured, and use it to write/read data, get statistics, and so on. Naturally, all of this would require a decent amount of documentation and testing, too. Once we have all of this worked out, the rest of this project (probably to be handled in other tickets) would be identifying what other backends we'd want in tahoe-lafs, then documenting, implementing, and testing them. We already have Amazon S3 and Rackspace as targets -- users of tahoe-lafs will probably have their own suggestions, and more backends will come up with more research.

[zooko]

See ​the RAIC diagram.

[kevan]

(this is an email I sent to zooko a while ago with my thoughts on how this should be implemented:)

First, I'll summarize, to make sure that I understand what you had in mind. Please correct me if you disagree with any of this.

The "redundant array of inexpensive clouds" idea means extending the current storage server in tahoe-lafs to support storage backends that aren't what we have now (writing shares to the local filesystem). Well actually, the redundant array of inexpensive clouds idea means doing that, then implementing plugins for popular existing cloud storage services -- Amazon S3 and Rackspace are two that you've mentioned, but there are probably others (if we end up going through with this, I'll probably email tahoe-dev so I can get an idea of what else is out there/what else people want to see supported, in addition to my own research).

The benefit (or at least the benefit that seems clear to me from your explanation -- perhaps there are others that are more obvious if you run a big tahoe-lafs installation like allmydata.com, or if you're more familiar with tahoe-lafs than I am) is decoupling the ability of a tahoe-lafs node to store files from its physical filesystem. So if, say, allmydata.com were to start running tahoe-lafs nodes using S3 as a backend, and their grid was filled, they could create more space on the grid by buying more S3 buckets, rather than upgrading physical servers or adding new servers (I've never used S3, but I would bet that it is easier to buy more S3 buckets than to upgrade servers). Or, if you wanted to create a grid without purchasing a bunch of servers, you could run a bunch of nodes on one machine (I was thinking vmware images, but then I started wondering whether it was even necessary to have that level of separation between tahoe-lafs nodes -- is it? but that's not really on topic), each mapping to a different S3 bucket or buckets.

Am I missing anything (aside from more examples)?

It seems like -- at least for S3 -- you could already sort of do this. There are projects like s3fs, which provide a FUSE interface to an S3 bucket (though the last file for it is more than a year old. it seems like there should be other projects like that, though) (edit: this is actually wrong -- I just hadn't found the Google code project, which is at ​http://code.google.com/p/s3fs/). Using that, you could mount your S3 bucket somewhere in the filesystem of your server, then kajigger the basedir of the tahoe-lafs node so that it rests in that area of the filesystem, or otherwise configure the tahoe-lafs node to save files there. This requires more work than what we'd eventually want with "redundant array of inexpensive clouds", of course, and (depending on how well FUSE or other S3 interfaces play) may only work on tahoe-lafs nodes running one unix or other, but if an operator got it working, it seems like they'd have most of the benefit outlined above without any further work on my/our part.

(not that I mind working on this, of course, but I figured it would be worthwhile to mention that)

In any case, I think implementing this would come down to two basic parts.

The first part would be adapting the existing codebase to use multiple backends.

We already have one backend -- the filesystem backend -- which I think should be a plugin in the same sense that the others will be plugins (i.e.: other code in tahoe-lafs can interact with a filesystem plugin without caring very much about how or where it is storing its files -- otherwise it doesn't seem very extensible). If you accept this, then we'd need to figure out what a backend plugin should look like. Maybe we can make each plugin implement RIStorageServer, and leave it at that. Then we might not need to do very much work on the existing server to make it work with the rest of the (new) system. However, it's possible that there is backend-independent logic in the current server implementation that we wouldn't want to duplicate in every other backend implementation. To address this, we could instead make a sort of backend-agnostic storage server that implements RIStorageServer, then make another interface for backends to implement, say IStorageProvider. The skeletal RIStorageServer would instantiate its IStorageProvider based on what the user configured, and use it to write/read data, get statistics, and so on. Then IStorageProvider would be a fairly simplistic filesystem-ish API.

The other part of preparation would be figuring out how to map user configuration choices to what actually happens when a node is started. Also, we'd want to figure out how (if?) we need to do anything special with the credentials that users might need to log in to their storage backend. I'll have a better idea of how I'd implement this once I look at the way it works for other things that users configure.

Naturally, all of this would require a decent amount of documentation and testing, too.

(I'm open to other ideas, of course -- these are just what came to my mind)

Once we have all of this worked out, the rest of this project would be identifying what other backends we'd want in tahoe-lafs, then documenting, implementing, and testing those. We already have Amazon S3 and Rackspace as targets -- users of tahoe-lafs will probably have their own suggestions, and more backends will come up with more research.

[davidsarah]

Generalizing this to include support for multiple backends (since I don't think we want to do it in a way that would only support S3 and local disk).

[davidsarah]

fix typo

[davidsarah]

Update description to reflect kevan's suggested approach.

[zooko]

Here is an incomplete patch for others (arc) to look at or improve.

[arch_o_median]

Implements tests of read and write for the nullbackend

[arch_o_median]

just so I don't lose it all...

[arch_o_median]

another checkpoint

[arch_o_median]

more precise tests in TestServerFSBackend

[arch_o_median]

backing myself up, some comments cleaned in interfaces, new tests in test_backends

[arch_o_median]

tiny change, now tests that allocated returns correct value

[arch_o_median]

The null backend test is useful for testing what happens when there's no effective limit on the backend

[arch_o_median]

checkpoint 9

[arch_o_median]

Completed coverage of remote_allocate_buckets

[arch_o_median]

(JACP) Just Another CheckPoint?

[arch_o_median]

all storage_index (word tokens) to storageindex in storage/server.py

[arch_o_median]

no longer trying to mock FS in TestServerFSBackend

[arch_o_median]

OK jacp15 contains a test that (almost) completely covers remote_allocate_buckets with the new backend. We should review this patches contents before writing more tests.

[davidsarah]

I'll review this.

[davidsarah]

Before going much further in relying on twisted.python.filepath.FilePath, can we think about the Unicode issue raised in ticket:1437#comment:3? Currently, storage directories with Unicode paths are intended to be supported on Windows.

[arch_o_median]

Replying to davidsarah:

Before going much further in relying on twisted.python.filepath.FilePath, can we think about the Unicode issue raised in ticket:1437#comment:3? Currently, storage directories with Unicode paths are intended to be supported on Windows.

OK... I guess that I should look into the twisted project's testing framework to determine what they know about this issue...

I'm currently snooping for leads here: ​http://twistedmatrix.com/trac/ticket/4736

[arch_o_median]

Replying to arch_o_median:

Replying to davidsarah:

Before going much further in relying on twisted.python.filepath.FilePath, can we think about the Unicode issue raised in ticket:1437#comment:3? Currently, storage directories with Unicode paths are intended to be supported on Windows.

OK... I guess that I should look into the twisted project's testing framework to determine what they know about this issue...

I'm currently snooping for leads here: ​http://twistedmatrix.com/trac/ticket/4736

So it seems like there may be (but probably there is not) an issue regarding Windows path representations to users versus to "OS" API's snooping here:

​http://twistedmatrix.com/trac/ticket/2366

[arch_o_median]

Replying to arch_o_median:

Replying to arch_o_median:

Replying to davidsarah:

Before going much further in relying on twisted.python.filepath.FilePath, can we think about the Unicode issue raised in ticket:1437#comment:3? Currently, storage directories with Unicode paths are intended to be supported on Windows.

OK... I guess that I should look into the twisted project's testing framework to determine what they know about this issue...

I'm currently snooping for leads here: ​http://twistedmatrix.com/trac/ticket/4736

So it seems like there may be (but probably there is not) an issue regarding Windows path representations to users versus to "OS" API's snooping here:

​http://twistedmatrix.com/trac/ticket/2366

(Is replying to myself bad form?) OK so I can't tell how 2366 is (or is not resolved) should I get a twisted login so I can ask about it on that ticket... I await direction.

[zooko]

I did some investigation about non-ASCII filename handling in filepath and in Tahoe-LAFS and posted my notes on ​Twisted #5203.

[arch_o_median]

After some chatting with zooko and warner in IRC, I've tentatively decided to use composition to inform the base Crawler object about the backend it is associated with. I'm not sure, but I think passing the whole <backend>Core object might be appropriate.

[Zancas]

My current test suite contains several tests that Zooko calls "transparent box". I need to decide whether they are appropriate:

1. remote_allocate_buckets populates incoming with shnum(s)
2. an attempt to allocate the same share (same ss) does _not_ create a new bucketwriter
3. test allocated size
4. together remote_write, remote_close, get_shares, and read_share_data behave

Since I am altering the location (from server to backend/core) of some of this functionality, and since I am altering the mechanism by which the filesystem is manipulated (to FilePath)... I think all of these tests are necessary.

It would be nice if the tests were designed to ensure the proper behavior independent of the underlying storage medium... but I think I need to assume a filesystem-like interface for at least (1,2, and 4), probably (3) as well...

[Zancas]

I'm confused about leases. When I look at the constructor for an immutable share file in a 'pristine' repository, (or in my latest version for that matter) I see that in the "create" clause of the constructor a python string representation of a big endian '0' is used for the number of leases.

​http://tahoe-lafs.org/trac/tahoe-lafs/browser/trunk/src/allmydata/storage/immutable.py#L63

This is confusing because in my test vector data (created some time ago) I have '1' as the initial number of leases. My guess is that I somehow got a bum test-vector value, but it'd be nice to hear from an architect that immutable share files really should start life with '0' leases!

[Zancas]

Patch passes allmydata.test.test_backends.TestServerAndFSBackend.test_write_and_read_share

[zooko]

Cool! Will review.

[Zancas]

TestServerAndFSBackend.test_read_old_share passes

[Zancas]

TestServerAndFSBackend passes all (3) tests

[Zancas]

5 test_backend tests pass

[Zancas]

uggg... bugs...

[Zancas]

the 5 tests pass... so what?

[zancas]

Ticket ​1465 more succinctly organizes the same code contained in these patches.

[zooko]

I added attachment:backends-configuration-docs.darcs.patch​ which contains documentation of the configuration options for the backends feature. I like Brian Warner's approach to development where he writes the docs first, even before the tests. (He writes tests second.) I encourage anyone working on this ticket to read (and possibly improve/fix/extend) these docs!

[davidsarah]

Review of backends-configuration-docs.darcs.patch​:

s3.rst:

• Add a short introduction saying what S3 is and why anyone might want to use it.

• It's a bit inconsistent that the value of the backend option is uppercase "S3", but the other option names are lowercase "s3_*". Also, I would make it "s3.*", since that's similar to the use of "." to group other related options.

• Should the s3_url option include the scheme name, i.e. defaulting to ​http://s3.amazonaws.com ? We might want to support https in future (although there would be more to configure if we check certificates).

• In the description of s3_max_space, copy the paragraph starting "This string contains a number" from disk.rst rather than referring to it.

• "enabling ``s3_max_space`` causes an extra S3 usage query to be sent for each share upload, causing the upload process to run slightly slower and incur more S3 request charges."

Each space query could be amortized over several uploads, using an estimate of the used space in-between. (That wouldn't be accurate if there are several storage servers accessing the same bucket, but it would be accurate enough if the maximum number of such servers is limited.) Even if we don't implement that right away, I'm not sure that this performance issue needs to go in s3.rst.

disk.rst:

• "Storing Shares in local filesystem" -> "Storing Shares on a Local Filesystem"

• use backend = disk, and say that it is the default.

configuration.rst:

• "Clients will be unaware of what backend is used by the server." -> "Clients need not be aware of which backend is used by a server."

• "including how to limit the space that will be consumed" -> "including how to reserve a minimum amount of free space"

[zooko]

I closed the subsidiary ticket #1465 as "fixed". The current patch set for this ticket as of this writing is attachment:20110829passespyflakes.darcs.patch​ (from that ticket) plus attachment:backends-configuration-docs.darcs.patch​.

[davidsarah]

This is just a "flat" recording of my refactoring of pluggable backends. I'll do a better recording tomorrow, and explain the refactoring.

[davidsarah]

This is still just a flat recording (a lot more changes to tests were needed than I anticipated).

[davidsarah]

Bleeding edge pluggable backends code from David-Sarah. refs #999

[davidsarah]

Rerecording of pluggable-backends-davidsarah-v3.darcs.patch that should fix the darcs performance problem when applied to trunk.

[davidsarah]

Work-in-progress, includes fix to bug involving BucketWriter?. refs #999

[zancas]

Replying to davidsarah:

Review of backends-configuration-docs.darcs.patch​:

s3.rst:

• Add a short introduction saying what S3 is and why anyone might want to use it.

• It's a bit inconsistent that the value of the backend option is uppercase "S3", but the other option names are lowercase "s3_*". Also, I would make it "s3.*", since that's similar to the use of "." to group other related options.

• Should the s3_url option include the scheme name, i.e. defaulting to http://s3.amazonaws.com ? We might want to support https in future (although there would be more to configure if we check certificates).

• In the description of s3_max_space, copy the paragraph starting "This string contains a number" from disk.rst rather than referring to it.

• "enabling ``s3_max_space`` causes an extra S3 usage query to be sent for each share upload, causing the upload process to run slightly slower and incur more S3 request charges."

Each space query could be amortized over several uploads, using an estimate of the used space in-between. (That wouldn't be accurate if there are several storage servers accessing the same bucket, but it would be accurate enough if the maximum number of such servers is limited.) Even if we don't implement that right away, I'm not sure that this performance issue needs to go in s3.rst.

disk.rst:

• "Storing Shares in local filesystem" -> "Storing Shares on a Local Filesystem"

• use backend = disk, not backend = local filesystem, and say that it is the default.

configuration.rst:

• "Clients will be unaware of what backend is used by the server." -> "Clients need not be aware of which backend is used by a server."

• "including how to limit the space that will be consumed" -> "including how to reserve a minimum amount of free space"

• currently clients _are_ aware of backend type.

[davidsarah]

docs: document the configuration options for the new backends scheme. This takes into account ticket:999#comment:26 and is rerecorded to avoid darcs context problems.

[zooko]

Replying to zancas:

• currently clients _are_ aware of backend type.

They are? I don't think so. How would they find out about the backend type?

[zooko]

attachment:backends-configuration-docs-v2.darcs.patch​ looks good to me. One thing I would change is to remove the "Issues" section about the costs of querying S3 objects and the effects on our crawler/lease-renewal scheme. I'm not sure that this branch will eventually land without a lease-checker implemented, so that part is making a statement that might be wrong. Also I'm not really sure the costs of querying S3 objects are worth mentioning. The ​current S3 pricing has 10,000 GET requests for $0.01. Let's remove that documentation for now and add in documentation when we understand better what the actual limitations or costs will be.

[davidsarah]

Replying to zancas:

Replying to davidsarah:

configuration.rst:

• "Clients will be unaware of what backend is used by the server." -> "Clients need not be aware of which backend is used by a server."

• currently clients _are_ aware of backend type.

The doc meant that client nodes need not be aware of backend type. Although the current hack to wire up a StorageServer to a backend in pluggable-backends-davidsarah-v5.darcs.patch​ is in allmydata/client.py, that code isn't actually run by clients, it is run only when setting up a storage server.

[davidsarah]

Replying to davidsarah:

Replying to zancas:

Replying to davidsarah:

configuration.rst:

• "Clients will be unaware of what backend is used by the server." -> "Clients need not be aware of which backend is used by a server."

• currently clients _are_ aware of backend type.

The doc meant that client nodes need not be aware of backend type.

Ugh, I should never use the term "node" :-/. I meant the code that acts as a storage protocol client.

[davidsarah]

v6. Tests are looking in much better shape now -- still some problems with path vs FilePath? and other stale assumptions in the test framework, but the disk backend basically works now.

[davidsarah]

Add --trace-exceptions option to trace raised exceptions on stderr. refs #999

[davidsarah]

Latest snapshot, more tests passing.

[zooko]

snapshot of work in progress

[davidsarah]

v8 snapshot. More tests pass.

[davidsarah]

Still more test fixes.

[davidsarah]

Josh wrote, re: pluggable-backends-davidsarah-v8.darcs.patch​:

I think the test_crawlers failure stems from ShareCrawler being passed a FilePath object in its constructor where it expects a string literal to use in an old-style call to open (specifically in its "load_state" method). I'm not certain yet, but I think I'll stop here for the night.

No, load_state uses pickle.loads(self.statefp.getContent()) which is correct. The state handling is a red herring for the test_crawlers failure, I think.

[davidsarah]

In v9, allmydata.test.test_storage.LeaseCrawler.test_basic is hanging due to an infinite recursion in pickle.py. Use

bin/tahoe --trace-exceptions debug trial --rterror allmydata.test.test_storage.LeaseCrawler.test_basic

(with trace-exceptions-option.darcs.patch​ applied) to see the recursion. I'm on the case...

[davidsarah]

Replying to davidsarah:

In v9, allmydata.test.test_storage.LeaseCrawler.test_basic is hanging due to an infinite recursion in pickle.py.

That was another red herring; there was an innocuous exception in pickle.py that was happening in each iteration of whatever other code is livelocking.

[davidsarah]

Fix most of the crawler tests. Reinstate the cancel_lease methods of ImmutableDiskShare? and MutableDiskShare?, since they are needed for lease expiry. refs #999

[davidsarah]

Includes a fix for iterating over a dict while removing entries from it in mutable/publish.py, some cosmetic changes, and a start on the S3 backend.

[davidsarah]

Updates to null and S3 backends.

[zancas]

contains changes in v12

[davidsarah]

Includes fixes to test_status_bad_disk_stats and test_no_st_blocks in test_storage.py, and more work on the S3 backend.

[davidsarah]

Work in progress for asyncifying the backend interface (necessary to call txaws methods that return Deferreds). This is incomplete so lots of tests fail. refs #999

[davidsarah]

In v13, test_storage.LeaseCrawler.test_share_corruption fails. However this is a test that is known to have race conditions -- it used to fail when logging was enabled (#923), and we tried to fix that in 3b1b0147a867759c, but in a way that in retrospect didn't really address the cause of the race condition. The problem is that it's trying to check for a particular instantaneous state of the lease crawler while it is running, which is inherently race-prone.

I suggest we not worry about this test for the current LAE iteration.

[davidsarah]

This does not include the asyncification changes from v14, but does include a couple of fixes for failures in test_system.

[davidsarah]

bleeding edge of asyncification work

[zancas]

Huh... weird, I can't apply v15...

0 /home/arc/sandbox/working 550 $ darcs apply pluggable-backends-davidsarah-v15.darcs.patch

darcs failed: Bad patch bundle! 2 /home/arc/sandbox/working 551 $

[davidsarah]

Latest asyncified patch. About 90% of tests pass.

[davidsarah]

Differences, just in the S3 backend, between v13a and v16.

[davidsarah]

Completes the splitting of IStoredShare into IShareForReading and IShareForWriting. Does not include configuration changes.

[davidsarah]

Snapshot of test_backends.py in David-Sarah's tree

[davidsarah]

Includes backend configuration (rerecorded from zooko's patch), and other minor fixes.

[davidsarah]

Include missing files for real and mock S3 backends. Also some fixes to tests, scripts/debug.py, and config parsing.

[david-sarah@…]

In [5373/ticket999-S3-backend]:

test_storage.py: only run test_large_share on the disk backend. (It will wedge your machine if run on the S3 backend with MockS3Bucket.) refs #999

[david-sarah@…]

In [5374/ticket999-S3-backend]:

test/mock_s3.py: fix a typo. refs #999

[david-sarah@…]

In [5375/ticket999-S3-backend]:

Make sure that the statedir is created before trying to use it. refs #999

[david-sarah@…]

In [5376/ticket999-S3-backend]:

s3_bucket.py: fix an incorrect argument signature for list_objects. refs #999

[david-sarah@…]

In [5379/ticket999-S3-backend]:

mock_s3.py: fix bug in MockS3Error constructor. refs #999

[david-sarah@…]

In [5380/ticket999-S3-backend]:

test_storage.py: Server class uses ShouldFailMixin?. refs #999

[david-sarah@…]

In [5382/ticket999-S3-backend]:

Add dummy lease methods to immutable S3 share objects. refs #999

[david-sarah@…]

In [5387/ticket999-S3-backend]:

s3/immutable.py: minor simplification in ImmutableS3ShareForReading. refs #999

[david-sarah@…]

In [5388/ticket999-S3-backend]:

Add a share._get_filepath() method used by tests to get the FilePath? for a share, rather than accessing the _home attribute. refs #999

[david-sarah@…]

In [5391/ticket999-S3-backend]:

s3/s3_common.py: remove incorrect 'self' arguments from interface methods in IS3Bucket. refs #999

[david-sarah@…]

In [5392/ticket999-S3-backend]:

More asyncification of tests. Also fix some bugs due to capture of slots in for loops. refs #999

[davidsarah]

Fix various bugs and tests. v20

[davidsarah]

Re: pluggable-backends-davidsarah-v20.darcs.patch​, I made a mistake in recording it that will cause a conflict with the ticket999-S3-backend branch. I'll attach a fixed version.

[david-sarah@…]

In [5400/ticket999-S3-backend]:

Add a get_share method to IShareSet, to get a specific share. refs #999

[david-sarah@…]

In [5402/ticket999-S3-backend]:

Add a _get_sharedir() method on IShareSet, implemented by the disk and mock S3 backends, for use by tests. refs #999

[david-sarah@…]

In [5403/ticket999-S3-backend]:

Fix some miscapture bugs. refs #999

[david-sarah@…]

In [5404/ticket999-S3-backend]:

Fix a duplicate umid. refs #999

[david-sarah@…]

In [5405/ticket999-S3-backend]:

Remove unused load method and _loaded attribute from s3/mutable.py. refs #999

[david-sarah@…]

In [5406/ticket999-S3-backend]:

Remove an inapplicable comment. refs #999

[david-sarah@…]

In [5407/ticket999-S3-backend]:

Make sure that get_size etc. work correctly on an ImmutableS3ShareForWriting after it has been closed. Also simplify by removing the _end_offset attribute. refs #999

[david-sarah@…]

In [5408/ticket999-S3-backend]:

unlink() on share objects should be idempotent. refs #999

[david-sarah@…]

In [5409/ticket999-S3-backend]:

Partially asyncify crawlers. refs #999

[david-sarah@…]

In [5410/ticket999-S3-backend]:

More miscapture fixes. refs #999

[david-sarah@…]

In [5411/ticket999-S3-backend]:

Ensure that helper classes are not treated as test cases. Also fix a missing mixin. refs #999

[david-sarah@…]

In [5412/ticket999-S3-backend]:

disk backend: size methods should no longer return Deferreds. refs #999

[david-sarah@…]

In [5414/ticket999-S3-backend]:

test_storage.py: fix a trivial bug in LeaseCrawler?.test_unpredictable_future. refs #999

[davidsarah]

Please ignore pluggable-backends-davidsarah-v20.darcs.patch​; the equivalent of that patch is on the ticket999-S3-backend branch now.

[david-sarah@…]

In [5415/ticket999-S3-backend]:

storage/backends/disk/mutable.py: put back a correct assertion that had been disabled. storage/base.py: fix the bug that was causing that assertion to fail. refs #999

[davidsarah]

[5415/ticket999-S3-backend] fixes all but one of the tests in test_mutable.py.

[david-sarah@…]

In [5416/ticket999-S3-backend]:

test_storage.py: move some tests that were not applicable to all backends out of ServerTest?. refs #999

[david-sarah@…]

In [5417/ticket999-S3-backend]:

Instrument some assertions to report the failed values. refs #999

[david-sarah@…]

In [5419/ticket999-S3-backend]:

interfaces.py: resolve conflicts with trunk. refs #999

[david-sarah@…]

In [5421/ticket999-S3-backend]:

interfaces.py: resolve another conflict with trunk. refs #999

[david-sarah@…]

In [5422/ticket999-S3-backend]:

test_download.py: fix test_download_failover (it should tolerate non-existing shares in _clobber_most_shares). refs #999

[david-sarah@…]

In [5423/ticket999-S3-backend]:

Null backend: implement unlink and readv more correctly. refs #999

[david-sarah@…]

In [5424/ticket999-S3-backend]:

Make unlink() on share objects consistently idempotent. refs #999

[david-sarah@…]

In [5425/ticket999-S3-backend]:

S3 backend: move the implementation of list_objects from s3_bucket.py to s3_common.py, making s3_bucket.py simpler and list_objects easier to test independently. refs #999

[david-sarah@…]

In [5426/ticket999-S3-backend]:

Add fileutil.fp_list(fp) which is like fp.children(), but returns [] in case of a directory that does not exist. Use it to simplify the disk backend and mock S3 bucket implementations. refs #999

[david-sarah@…]

In [5427/ticket999-S3-backend]:

test/mock_s3.py: fix a bug that was causing us to use the wrong directory for share files. refs #999

[david-sarah@…]

In [5429/ticket999-S3-backend]:

test_storage.py: make MutableServer?.test_leases pass. refs #999

[david-sarah@…]

In [5430/ticket999-S3-backend]:

test_storage.py: fix a bug introduced by asyncification of test_allocate. refs #999

[david-sarah@…]

In [5431/ticket999-S3-backend]:

test_storage.py: fix a typo in test_null_backend. refs #999

[david-sarah@…]

In [5432/ticket999-S3-backend]:

test_storage.py: fix a trivial bug in MDMFProxies.test_write. refs #999

[david-sarah@…]

In [5433/ticket999-S3-backend]:

test_storage.py: fix asyncification of three tests in MDMFProxies. refs #999

[david-sarah@…]

In [5434/ticket999-S3-backend]:

Fix two pyflakes warnings about unused imports. refs #999

[david-sarah@…]

In [5445/ticket999-S3-backend]:

test_storage.py: cosmetics. refs #999

[david-sarah@…]

In [5446/ticket999-S3-backend]:

test_storage.py: fix test failures in MDMFProxies. refs #999

[david-sarah@…]

In [5447/ticket999-S3-backend]:

Move configuration of each backend into the backend itself. refs #999

[david-sarah@…]

In [5448/ticket999-S3-backend]:

util/deferredutil.py: remove unneeded utility functions. refs #999

[david-sarah@…]

In [5449/ticket999-S3-backend]:

test_storage.py: Move test_seek to its own class, since it is independent of the backend. Also move test_reserved_space to ServerWithDiskBackend?, since reserved_space is specific to that backend. refs #999

[david-sarah@…]

In [5450/ticket999-S3-backend]:

test_storage.py: add a test that we can create a share, exercising the backend's get_share and get_shares methods. This may explicate particular kinds of backend failure better than the existing tests. refs #999

[david-sarah@…]

In [5451/ticket999-S3-backend]:

test_storage.py: asyncify some more tests, and fix create methods. refs #999

[david-sarah@…]

In [5452/ticket999-S3-backend]:

S3 backend: fix corruption advisories and listing of shares for mock S3 bucket. refs #999

[david-sarah@…]

In [5453/ticket999-S3-backend]:

no_network.py: fix delete_all_shares. refs #999

[david-sarah@…]

In [5454/ticket999-S3-backend]:

test_download.py: fix and reenable Corruption.test_each_byte. Add a comment noting that catalog_detection = True has bitrotted. refs #999

[david-sarah@…]

In [5455/ticket999-S3-backend]:

test_storage.py: add test_write_and_read_share and test_read_old_share originally from test_backends.py. refs #999

[david-sarah@…]

In [5456/ticket999-S3-backend]:

Remove test_backends.py, since all its tests are now redundant with tests in test_storage.py or test_client.py. refs #999

[david-sarah@…]

In [5457/ticket999-S3-backend]:

Null backend: make NullShareSet? inherit from ShareSet?, which should implement readv correctly. Remove its implementation of testv_and_readv_and_writev since the one from ShareSet? should work (if it doesn't that would be a separate bug). refs #999

[david-sarah@…]

In [5458/ticket999-S3-backend]:

S3 backend: correct list_objects to list_all_objects in IS3Bucket. refs #999

[david-sarah@…]

In [5459/ticket999-S3-backend]:

storage/backends/base.py: allow readv to work for both mutable and immutable shares. refs #999

[david-sarah@…]

In [5461/ticket999-S3-backend]:

test_storage.py: test_read_old_share and test_write_and_read_share should only expect to be able to read input share data. refs #999

[david-sarah@…]

In [5462/ticket999-S3-backend]:

S3 backend: keep track of incoming shares, so that the storage server can avoid creating BucketWriters? for shnums that have an incoming share. refs #999

[david-sarah@…]

In [5463/ticket999-S3-backend]:

docs/backends/S3.rst: note that storage servers should use different buckets. refs #999

[david-sarah@…]

In [5464/ticket999-S3-backend]:

test_storage.py: fix a typo (d vs d2) in test_remove_incoming. refs #999

[david-sarah@…]

In [5465/ticket999-S3-backend]:

test_storage: rename the two test_leases methods to ServerTest?.test_immutable_leases and MutableServer?.test_mutable_leases. refs #999

[david-sarah@…]

In [5466/ticket999-S3-backend]:

test_storage: fix some typos introduced when asyncifying test_immutable_leases. refs #999

[david-sarah@…]

In [5467/ticket999-S3-backend]:

test_storage: in test_no_st_blocks, print the rec 'dict' if checking one of its fields fails. refs #999

[david-sarah@…]

In [5468/ticket999-S3-backend]:

test_storage.py: remove some redundant coercions to bool. refs #999

[david-sarah@…]

In [5469/ticket999-S3-backend]:

test_storage.py: print more info when checks fail. refs #999

[david-sarah@…]

In [5470/ticket999-S3-backend]:

test_storage.py: fix two bugs in test_no_st_blocks -- the _cleanup function was being called too early, and we needed to treat directories as using no space in order for the configured-sharebytes == configured-diskbytes check to be correct. refs #999

[david-sarah@…]

In [5471/ticket999-S3-backend]:

Undo partial asyncification of crawlers, and enable crawlers only for the disk backend. refs #999

[david-sarah@…]

In [5472/ticket999-S3-backend]:

test_storage.py: fix a bug in _backdate_leases (it was returning too early). refs #999

[david-sarah@…]

In [5473/ticket999-S3-backend]:

scripts/debug.py: fix stale code in describe_share that had not been updated for changes in share interfaces. refs #999

[david-sarah@…]

In [5474/ticket999-S3-backend]:

Disk backend: make sure that disk shares with a storageindex of None (as sometimes created by test code) can be printed using repr. refs #999

[david-sarah@…]

In [5475/ticket999-S3-backend]:

Change accesses of ._sharehomedir on a disk shareset to _get_sharedir(). refs #999

[david-sarah@…]

In [5476/ticket999-S3-backend]:

test_storage.py: cleanup to style of test_limited_history to match other tests. refs #999

[david-sarah@…]

In [5477/ticket999-S3-backend]:

Change IShareSet.get_shares[_synchronous] to return a pair (list of share objects, set of corrupt shnums). This is necessary to allow crawlers to record but skip over corrupt shares. This patch also changes the behaviour of storage servers to ignore corrupt shares on read, which may or may not be what we want. Note that the S3 backend does not yet report corrupt shares. refs #999

[david-sarah@…]

In [5478/ticket999-S3-backend]:

Allow crawlers and storage servers to use a deterministic clock, for testing. We do not yet take advantage of this in tests. refs #999

[david-sarah@…]

In [5479/ticket999-S3-backend]:

Fix race conditions in crawler tests. (storage.LeaseCrawler?.test_unpredictable_future may still be racy.) refs #999

[david-sarah@…]

In [5480/ticket999-S3-backend]:

Add some repr methods. refs #999

[davidsarah]

In [5479/ticket999-S3-backend], there's also a fix to a preexisting bug in test_storage.LeaseCrawler.test_unpredictable_future, where it was checking the s["estimated-remaining-cycle"]["space-recovered"] key twice, rather than both that key and s["estimated-current-cycle"]["space-recovered"] as intended.

[david-sarah@…]

In [5481/ticket999-S3-backend]:

test_storage.py, test_crawler.py: change 'bucket' terminology to 'shareset' where appropriate. refs #999

[david-sarah@…]

In [5482/ticket999-S3-backend]:

S3 backend: remove max_space option. refs #999

[david-sarah@…]

In [5483/ticket999-S3-backend]:

Enable mutable tests for S3 backend (they all fail, as expected). refs #999

[david-sarah@…]

In [5484/ticket999-S3-backend]:

storage/backends/disk/mutable.py: correct a typo. refs #999

[david-sarah@…]

In [5485/ticket999-S3-backend]:

Disk backend: fix incorrect arguments in a call to create_mutable_disk_share. refs #999

[david-sarah@…]

In [5486/ticket999-S3-backend]:

test_storage.py: move the test_container_size test to MutableServerWithDiskBackend? for now, because it tries to create a very large container which will wedge your machine. refs #999

[david-sarah@…]

In [5487/ticket999-S3-backend]:

S3 backend: finish implementation of mutable shares. refs #999

[david-sarah@…]

In [5488/ticket999-S3-backend]:

test_storage.py: reduce duplicated code by factoring 'create' methods into CreateS3Backend and CreateDiskBackend? classes. refs #999

[david-sarah@…]

In [5489/ticket999-S3-backend]:

S3 backend: make sure that the container size limit is checked before writing. refs #999

[david-sarah@…]

In [5490/ticket999-S3-backend]:

S3 backend: make precondition failures show more information. refs #999

[david-sarah@…]

In [5491/ticket999-S3-backend]:

S3 backend: new_length argument to MutableS3Share.writev should only be able to truncate the share (after applying writes), not extend it. refs #999

[david-sarah@…]

In [5492/ticket999-S3-backend]:

S3 backend: the mutable size limit should be on the data length, not the container size. Also simplify by removing _check_size_limit. refs #999

[david-sarah@…]

In [5493/ticket999-S3-backend]:

Disk backend: make sure that the size limit is checked before writing. Also, the size limit is on the data length, not the container size. refs #999

[david-sarah@…]

In [5494/ticket999-S3-backend]:

test_storage.py: reenable MutableServer?.test_container_size for the S3 backend. refs #999

[david-sarah@…]

In [5495/ticket999-S3-backend]:

test_storage.py: the part of test_remove that checks non-existence of the share directory after deleting a share, is only applicable to the disk backend; but, we can check that the shareset has no overhead at that point. refs #999

[david-sarah@…]

In [5514/ticket999-S3-backend]:

Add a '[storage]backend = mock_s3' option for use by tests. Move mock_s3.py to src/allmydataa/storage/backends/s3 since it is now imported by non-test code. refs #999

[david-sarah@…]

In [5515/ticket999-S3-backend]:

test_system.py: enable system tests to run against S3 backend as well as disk backend. refs #999

[david-sarah@…]

In [5516/ticket999-S3-backend]:

test_system.py: fix a typo. refs #999

[david-sarah@…]

In [5517/ticket999-S3-backend]:

test_system.py: rename ServerTestWith?*Backend to ServerWith?*Backend, for consistency with tst_storage.py. refs #999

[david-sarah@…]

In [5518/ticket999-S3-backend]:

test_system.py: make checks in _test_runner more picky about field names to avoid accidental suffix matches. refs #999

[david-sarah@…]

In [5519/ticket999-S3-backend]:

test_system.py: ensure that subclasses of SystemTest? use different test directories. refs #999

[david-sarah@…]

In [5520/ticket999-S3-backend]:

test_system.py: fix SystemWithS3Backend.test_mutable by only requiring the line specifying which nodeid the lease secrets are for when the node has a disk backend. refs #999

[david-sarah@…]

In [5521/ticket999-S3-backend]:

scripts/debug.py: in catalog-shares, gracefully handle the case where a share has no leases (for example because it is an S3 share). refs #999

[david-sarah@…]

In [5522/ticket999-S3-backend]:

test_system.py: check that there is no error output from invocations of 'tahoe debug'. refs #999

[david-sarah@…]

In [5523/ticket999-S3-backend]:

mock_s3.py: remove bucketname argument to MockS3Bucket constructor, since it is not needed. refs #999

[david-sarah@…]

In [5524/ticket999-S3-backend]:

S3 backend: remove support for [storage]readonly option. refs #999, #1568

[david-sarah@…]

In [5525/ticket999-S3-backend]:

S3 backend: the s3.region option is unnecessary; it is only used for EC2 endpoints, and we only need an S3 one. Also simplify wording in S3.rst. refs #999

[david-sarah@…]

In [5526/ticket999-S3-backend]:

docs/backends/S3.rst: document the requirement for the storage server to have the correct time to within 15 minutes. refs #999

[davidsarah]

Further work on this functionality will be in ticket #1569.