Summary:
Ref T12613. Currently, the SVNTEST and HGTEST repositories are improperly configured on `secure`. These repositories use VCS systems which do not support synchronization, so they can not be served from cluster services with multiple hosts.
However, I've incorrectly configured them the same way as all the Git repositories, which support synchronization. This causes about 50% of requests to randomly fail (when they reach the wrong host).
Detect this issue and warn the user that the configuration is not valid.
It should be exceptionally difficult for normal installs to run into this.
Test Plan:
- Mostly faked these conditions locally, verified that `secure` really has this configuration.
- I'll push this, verify that the issue is detected correctly in production, then fix the config which should resolve the intermittent issues with SVNTEST.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12613
Differential Revision: https://secure.phabricator.com/D17774
Summary:
Via HackerOne (<https://hackerone.com/reports/220909>). When we close commits in response to "Fixes Txxx", we currently act as the omnipotent user. This allows users to close tasks they can't see by pushing commits with "Fixes Txxx" in the message.
However, we can't actually tell who authored or committed a change: we're just using the "Author" and "Committer" values from Git in most cases, and anyone can forge those. So we can't really get this right, in a security sense.
(We can tell who //pushed// a change if we host it, but that's often not the right user. If GPG signing was more prevalent, we could use that. In the future, we could use side channels like having `arc land` tell Phabrcator who was pushing changes.)
Since I think the impact of this is fairly minor and this isn't //really// a security issue (more of a confusion/abuse/product issue) I think the behavior is okay more-or-less as-is, but we can do better when we do identify an author: drop permissions, and use their privileges to load the tasks which the commit "fixes".
This effectively implements this rule:
> If we identify the author of a commit as user X, that commit can only affect tasks which user X can see and edit.
Note that:
- Commits which we can't identify the author for can still affect any task.
- Any user can forge any other user's identity (or an invalid identity) and affect any task.
So this is just a guard rail to prevent mistakes by good-faith users who type the wrong task IDs, not a real security measure.
Also note that to perform this "attack" you must already have commit access to a repository (or permission to create a repository).
Test Plan:
- Used `bin/repository reparse --message <commit> --force-autoclose` to run the relevant code.
- Made the code `throw` before it actually applied the edit.
- Verified that the edit was rejected if the author was recognized and can not see or could not edit the task.
- Verified that the edit is accepted if the author can see+edit the task.
- Verified that the edit is accepted if we can't figure out who the author is.
Reviewers: chad
Reviewed By: chad
Differential Revision: https://secure.phabricator.com/D17688
Summary:
Ref T12298. The PullLocal daemon has had hibernation code for a little while, but it never actually activated because we don't sleep for more than 15 seconds in any case.
Add a maximum sleep instead and use that to control the longest sleep we'll do for hibernation purposes.
Also, when a repository or repository URI is edited, write a NEEDS_UPDATE event into the message table to make sure the daemons de-hibernate.
Test Plan: Used `bin/phd debug pull`, saw the daemon actually hibernate instead of just sleeping for 15 seconds.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12298
Differential Revision: https://secure.phabricator.com/D17635
Summary:
This implements a simplistic `PhabricatorRepositoryFulltextEngine`
Currently only the repository name, description, timestamps and
status are indexed.
Note: I had to change the `search index` workflow to disambiguate
PhabricatorRepository from PhabricatorRepositoryCommit
Test Plan:
* ran `./bin/search index --type PhabricatorRepository --force`
* searched for some repositories. Saw reasonable results matching on either title or description.
* Edited a repository in the web ui
* Added unique key words to the repo description.
* I was then able to find that repo by searching for the new keywords.
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: Korvin
Tags: #search, #diffusion
Differential Revision: https://secure.phabricator.com/D17300
Summary: Ref T12298. This allows the PullLocal daemon to hibernate like the Trigger daemon, but automatically wakes it back up when it needs to do something.
Test Plan:
- Ran `bin/phd debug pulllocal --trace`.
- Saw the daemon hibernate after doing a checkup on repositories.
- Saw periodic queries to look for new update messages.
- After clicking "Update Now" in the web UI to schedule an update, saw the daemon wake up immediately.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12298
Differential Revision: https://secure.phabricator.com/D17540
Summary:
Ref T10967. Improves some method names:
- `Revision->getReviewerStatus()` -> `Revision->getReviewers()`
- `Revision->attachReviewerStatus()` -> `Revision->attachReviewers()`
- `Reviewer->getStatus()` -> `Reviewer->getReviewerStatus()` (this is mostly to make this more greppable)
Test Plan:
- bunch o' `grep`
- Browsed around.
- If I missed anything, it should fatal in an obvious way. We have a lot of other `getStatus()` calls and it's hard to be sure I got them all.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T10967
Differential Revision: https://secure.phabricator.com/D17522
Summary: Ref T10967. The old name was because we had a `getReviewers()` tied to `needRelationships()`, rename this method to use a simpler and more clear name.
Test Plan: `grep`, browsed around.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T10967
Differential Revision: https://secure.phabricator.com/D17519
Summary: Fixes T12416. See that task for discussion. Slightly older versions of `git` do not appear to support use of `--` to separate flags and arguments.
Test Plan:
- Ran `bin/repository update PHABX`.
- In T12416, had a user with Git 2.1.4 confirm that `git ls-remote X` worked while `git ls-remote -- X` failed.
- Read `git help ls-remote` to look for any kind of suspicious `--destroy-the-world` flags, didn't see any that made me uneasy.
Reviewers: chad, avivey
Reviewed By: avivey
Maniphest Tasks: T12416
Differential Revision: https://secure.phabricator.com/D17508
Summary:
Ref T12392. The logic currently goes like this:
- Try a fetch.
- If that fails, try repairing the origin URI.
- Then try again.
This is pretty complicated, and we can use this simpler logic instead:
- Set the origin URI to the right value.
- Try a fetch.
Setting the origin URI is very fast. This can normally only get us in any trouble in very obscure situations which haven't occurred for many years:
- Pretty much all of this is already covered by `verifyGitOrigin()`, which we run earlier.
- Origins could be configured to have multiple URIs for some reason, but shouldn't be.
- Years ago, you could configure Phabricator to point at a local repository it didn't own and that could conceivably have a different "origin" that you might not want us to delete. If you did this, the daemons have been spewing errors for 3-4 years without you fixing it. The cost of fixing the remote URI is very small even if anyone is affected by this (just set it back to the old value) and there's zero reason to do this and the scenario is ridiculous.
Test Plan: Ran `bin/repository update PHABX --trace --verbose`, saw fetches go through cleanly after URI adjustment.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12392
Differential Revision: https://secure.phabricator.com/D17498
Summary:
Ref T12296. Ref T12392. Currently, when we're observing a remote repository, we periodically run `git fetch ...`.
Instead, periodically run `git ls-remote` (to list refs in the remote) and `git for-each-ref` (to list local refs) and only continue if the two lists are different.
The motivations for this are:
- In T12296, it appears that doing this is //faster// than doing a no-op `git fetch`. This effect seems to reproduce locally in a clean environment (900ms for `ls-remote` + 100ms for `for-each-ref` vs about 1.4s for `fetch`). I don't have any explanation for why this is, but there it is. This isn't a huge change, although the time we're saving does appear to mostly be local CPU time, which is good for us.
- Because we control all writes, we could cache `git for-each-ref` in the future and do fewer disk operations. This doesn't necessarily seem too valuable, though.
- This allows us to tell if a fetch will do anything or not, and make better decisions around clustering (in particular, simplify how observed repository versioning works). With `git fetch`, we can't easily distinguish between "fetch, but nothing changed" and "legitimate fetch".
If a repository updates very regularly we end up doing slightly more work this way (that is, if `ls-remote` always comes back with changes, we do a little extra work), but this is normally very rare.
This might not get non-bare repositories quite right in some cases (i.e., incorrectly detect them as changed when they are unchanged) but we haven't created non-bare repositories for many years.
Test Plan: Ran `bin/repository update --trace --verbose PHABX`, saw sensible construction of local and remote maps and accurate detection of whether a fetch would do anything or not.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12392, T12296
Differential Revision: https://secure.phabricator.com/D17497
Summary:
Ref T12296. This cache is used to cache Git ref heads (branches, tags, etc). Reasonable repositories may have more than 2048 of these.
When we miss the cache, we need to single-get refs to check them, which is relatively expensive.
Increasing the size of the cache to 65535 should only require about 7.5MB of RAM.
Additionally, fill only as much of the cache as actually fits. The FIFO nature of the cache can get us into trouble otherwise.
If we insert "A, B, C, D" and then lookup A, B, C, D, but the cache has maximum size 3, we get this:
- Insert A, B, C, D: cache is now "B, C, D".
- Lookup A: miss, single get, insert, purge, cache is now "C, D, A".
- Lookup B: miss, singel get, insert, purge, cache is now "D, A, B".
Test Plan:
- Reduced cache size to 5, observed reasonable behavior on the `array_slice()` locally with `bin/repository update` + `var_dump()`.
- Used this script to estimate the size of 65535 cache entries as 7.5MB:
```
epriestley@orbital ~ $ cat size.php
<?php
$cache = array();
$mem_start = memory_get_usage();
for ($ii = 0; $ii < 65535; $ii++) {
$cache[sha1($ii)] = true;
}
echo number_format(memory_get_usage() - $mem_start)." bytes\n";
epriestley@orbital ~ $ php -f size.php
7,602,176 bytes
```
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12296
Differential Revision: https://secure.phabricator.com/D17409
Summary:
Ref T12173.
- If we want to fetch a tag, Buildkite needs it as a "branch" (this means more like "ref to fetch").
- The API gets upset if we pass "refs/tags/...", so just pass the tag name without the prefix, which works.
- Do a better job with commits and pass a real branch to fetch.
Test Plan:
- Built a commit with Buildkite.
- Build a revision with Buildkite.
Reviewers: chad
Reviewed By: chad
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T12173
Differential Revision: https://secure.phabricator.com/D17282
Summary:
Ref T10978. This updates audits triggered by Owners to use a modern transaction. Minor changes:
- After D17264, we no longer need the "AUDIT_NOT_REQUIRED" fake-audits to record package membership. This no longer creates them.
- This previously saved English-language, untranslatable text strings about audit details onto the audit relationship. I've removed them, per discussion in D17263.
The "Audit Reasons" here are potentially a little more useful than the Herald/Explicit-By-Owner ones were, since the rules are a little more complex, but I'd still like to see evidence that we need them.
In particular, the transaction record now says "Owners added auditors: ...", just like Differential, so the source of the auditors should be clear:
{F2549087}
T11118 (roughly "add several Owners audit modes", despite the title at time of writing) might impact this too. Basically, this is simple and maybe good enough; if it's not quite good enough we can refine it.
Test Plan: Ran `bin/repository reparse --owners <commit>` saw appropriate owners audits trigger.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T10978
Differential Revision: https://secure.phabricator.com/D17266
Summary:
Ref T10978. Currently, during commit import, we write an "Audit Not Required" auditor for commits which don't require an audit.
This auditor is used to power the "Commits in this package" query in Owners.
This conflates audits and commit/package membership. I think it might even predate edges. Code needs to dance around this mess and we get the wrong result in some cases, since auditors are now editable.
Instead, write an explicit edge which just says "this commit is part of such-and-such packages". Then use that to run the query. Logical!
I'll issue guidance on this but I'm not migrating it, since it fixes itself going forward and only really affects the UI in Owners.
Test Plan:
- Ran `bin/audit update-owners` with various arguments.
- Viewed packages in web UI, saw them load the proper commits.
- Queried by packages in Diffusion explicitly.
- Clicked the "View All" link in Owners and got to the right search UI.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T10978
Differential Revision: https://secure.phabricator.com/D17264
Summary:
Fixes T2393. This allows authors to explicitly say "I think I fixed everything, please accept my commit now thank you".
Also improves behavior of "re-accept" and "re-reject" after new auditors you have authority over get added.
Test Plan:
- Kicked a commit back and forth between an author and auditor by alternately using "Request Verification" and "Raise Concern".
- Verified it showed up properly in bucketing for both users.
- Accepted, added a project, accepted again (works now; didn't before).
- Audited on behalf of projects / packages.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T2393
Differential Revision: https://secure.phabricator.com/D17252
Summary:
Ref T2393. We had three copies of this code ("which packages/projects can a user accept on behalf of?"). I removed one in D17250. This consolidates the other two.
This still isn't perfect and it should probably live in a Query or something some day, but there's some weird stuff going on with the viewer in the editor context, and at least the code handles the viewer correctly now and isn't living somewhere weird and totally unrelated to auditing, and the callsites don't need to do a bunch of extra work.
This also moves towards fixing the "re-accept if you've already accepted but then a new package you have authority over was added" bug, which we fixed recently in Differential. This should be less common in Audit, but should still be fixed.
Test Plan: Viewed and audited commits with a mixture of user, package, and project auditors. Saw actions apply to the expected set of auditors.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T2393
Differential Revision: https://secure.phabricator.com/D17251
Summary: Fixes T5889. You can't write a rule like "if no other Herald rules did anything...", but you can use this rule to check for Owners or an explicit "Auditors" field doing things.
Test Plan: Using the test console, ran an "Auditors" rule against a commit with and without an auditor. Got expected pass/fail outcomes.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T5889
Differential Revision: https://secure.phabricator.com/D17221
Summary: Fixes T6660. Uses the new stuff in Audit to build an EditEngine-aware icon.
Test Plan: {F2364304}
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T6660
Differential Revision: https://secure.phabricator.com/D17208
Summary: Ref T10978. This is bare bones, but the SearchEngine is at least mostly in reasonable shape now, so get it in place and freeze the old stuff. I previously froze `audit.query`, which did much the same thing.
Test Plan: Issued some queries with the API, technically got results back.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T10978
Differential Revision: https://secure.phabricator.com/D17194
Summary:
Fixes T7076. This could probably use some tweaking but should get the basics in place.
This shows overall object state (e.g., "Needs Review"), not individual viewer state (e.g., "you need to review this"). After the bucketing changes it seems like we're mostly in a reasonable place on showing global state instead of viewer state. This makes the overall change much easier than it might otherwise have been.
Test Plan: {F2351867}
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T7076
Differential Revision: https://secure.phabricator.com/D17193
Summary: Fixes T7504. I think that task legitimately describes a bug and that the current behavior is counterintuitive.
Test Plan: Manually added an auditor to a commit with none; saw it become "Audit Required" as an overall state.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T7504
Differential Revision: https://secure.phabricator.com/D17185
Summary: Ref T10978. Ref T7676. Make auditors work more like reviewers, so they can be freely added or removed.
Test Plan:
- Interacted with auditors via "Edit Commit" and API.
- Comment area is still oldschool and doesn't work yet.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T10978, T7676
Differential Revision: https://secure.phabricator.com/D17181
Summary:
Fixes T12087. When transitioning into a clustered configuration for the first time, the documentation recommends using a one-device cluster as a transitional step.
However, installs may not do this for whatever reason, and we aren't as clear as we could be in warning about clusterizing directly into a multi-device cluster.
Roughly, when you do this, we end up believing that working copies exist on several different devices, but have no information about which copy or copies are up to date. //Usually// they all were already synchronized and are all up to date, but we can't make this assumption safely without risking data.
Instead, we err on the side of caution, and require a human to tell us which copy we should consider to be up-to-date, using `bin/repository thaw --promote`.
Test Plan:
```
$ ./bin/repository clusterize rLOCKS --service repos001.phacility.net
Service "repos001.phacility.net" is actively bound to more than one device
(local002.local, local001.phacility.net).
If you clusterize a repository onto this service it will be unclear which
devices have up-to-date copies of the repository. This leader/follower
ambiguity will freeze the repository. You may need to manually promote a
device to unfreeze it. See "Ambiguous Leaders" in the documentation for
discussion.
Continue anyway? [y/N]
```
Read other changes.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12087
Differential Revision: https://secure.phabricator.com/D17169
Summary:
Ref T12074. The "v3" API methods (`*.search`, `*.edit`) are currently marked as "unstable", but they're pretty stable and essentially all new code should be using them.
Although these methods are seeing some changes, almost all changes are additive (support for new constraints or attachemnts) and do not break backward compatibility. We have no major, compatibility-breaking changes planned.
I don't want to mark the older methods "deprecated" yet since `arc` still uses a lot of them and there are some capabilities not yet available on the v3 methods, but introduce a new "frozen" status with pointers to the new methods.
Overall, this should gently push users toward the newer methods.
Test Plan: {F2325323}
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12074
Differential Revision: https://secure.phabricator.com/D17158
Summary:
Fixes T9276. Fixes T8650. The story so far:
- We once published build updates to Revisions.
- An unrelated fix (D10911) sent them to the Diffs instead of Revisions, which isn't useful, since you can't see a diff's timeline anywhere.
- This also caused a race condition, where the RevisionEditor and DiffEditor would update the diff simultaneously (T8650).
- The diff update was just disabled to avoid the race (part of D13441).
- Instead, allow the updates to go somewhere else. In this case, we send commit updates to the commit but send diff updates to the revision so you can see 'em.
- Since everything will be using the revision editor now, we should either get proper lock behavior for free or it should be easy to add if something whack is still happening.
- Overall, this should pretty much put us back in working order like we were before D10911.
This behavior is undoubtedly refinable, but this should let us move forward.
Test Plan:
Saw a build failure in timeline:
{F2304575}
Reviewers: chad
Reviewed By: chad
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T9276, T8650
Differential Revision: https://secure.phabricator.com/D17139
Summary:
Fixes T12062. Like the commits from the year 3500, you can artificially build commits with no date information.
We could explicitly store these as `null` to fully respect the underlying datastore. However, I think it's very unlikely that these commits are intentional/meaningful or that this is valuable.
Additionally, "git show" interprets these commits as "Jan 1, 1970". Just store a `0` to mimic its behavior.
Test Plan:
- Following the process in T11537#192019, artificially created a commit with //no// date information (I deleted all date information from the message).
- Used `git show` / `git log --format ...` to inspect it: "Jan 1, 1970" on `git show`, no information at all on `%aD`, `%aT`, etc.
- Pushed it.
- Saw exception for trying to insert empty string into epoch colum from `bin/repository update`.
- Applied patch.
- Got a clean import.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T12062
Differential Revision: https://secure.phabricator.com/D17136
Summary:
Fetches cause output in `/var/tmp/phd/log/daemons.log` as
follows:
```
PHLOG: 'Unexpected output while updating repository "rREPONAME": No entry for terminal type "unknown";
using dumb terminal settings.
' at [/path/to/phabricator/src/applications/repository/daemon/PhabricatorRepositoryPullLocalDaemon.php:455]
```
These warnings come from PHP itself. Silence these warnings by providing a
known value for `TERM` before shelling out to the PHP script.
See also D9744 (reverted in D11644) and T4990/T7119, which are a similar issue,
but in the pre-receive hooks, not the pull daemons.
Test Plan:
Enabled in production, observed errors to be silenced and
no SSH hangs
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: epriestley
Differential Revision: https://secure.phabricator.com/D17100
Summary:
Ref T11954. This is kind of complex and I'm not sure I want to actually land it, but it gives us a fairly good improvement for clustered repositories so I'm leaning toward moving forward.
When we make (or receive) clustered repository requests, we must first load a bunch of stuff out of Almanac to figure out where to send the request (or if we can handle the request ourselves).
This involves several round trip queries into Almanac (service, device, interfaces, bindings, properties) and generally is fairly slow/expensive. The actual data we get out of it is just a list of URIs.
Caching this would be very easy, except that invalidating the cache is difficult, since editing any binding, property, interface, or device may invalidate the cache for indirectly connected services and repositories.
To address this, introduce `PhabricatorCacheEngine`, which is an extensible engine like `PhabricatorDestructionEngine` for propagating cache updates. It has two modes:
- Discover linked objects (that is: find related objects which may need to have caches invalidated).
- Invalidate caches (that is: nuke any caches which need to be nuked).
Both modes are extensible, so third-party code can build repository-dependent caches or whatever. This may be overkill but even if Almanac is the only thing we use it for it feels like a fairly clean solution to the problem.
With `CacheEngine`, make any edit to Almanac stuff propagate up to the Service, and then from the Service to any linked Repositories.
Once we hit repositories, invalidate their caches when Almanac changes.
Test Plan:
- Observed a 20-30ms performance improvement with `ab -n 100`.
- (The main page making Conduit calls also gets a performance improvement, although that's a little trickier to measure directly.)
- Added debugging code to the cache engine stuff to observe the linking and invalidation phases.
- Made invalidation throw; verified that editing properties, bindings, etc, properly invalidates the cache of any indirectly linked repositories.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T11954
Differential Revision: https://secure.phabricator.com/D17000
Summary: Ref T929. We've made some UI updates since D15330.
Test Plan: {F2079125}
Reviewers: avivey, chad
Reviewed By: chad
Maniphest Tasks: T929
Differential Revision: https://secure.phabricator.com/D16990
Summary:
Fixes T11940. In 2.11.0, Git has made a change so that newly-pushed changes are held in a temporary area until the hook accepts or rejects them.
This magic temporary area is only readable if the appropriate `GIT_ENVIRONMENTAL_MAGIC` variables are available. When executing `git` commands, pass them through from the calling context.
We're intentionally conservative about which variables we pass, and with good reason (see "httpoxy" in T11359). I think this continues to be the correct default behavior.
Test Plan:
- Upgraded to Git 2.11.0.
- Tried to push over SSH, got a hook error.
- Applied patch.
- Pulled and pushed over SSH.
- Pulled and pushed over HTTP.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T11940
Differential Revision: https://secure.phabricator.com/D16988
Summary:
Fixes T11902.
- Periods now work in short names.
- If you try to name something ".git", no dice.
Test Plan:
- Tried to name something "quack.git", was politely rejected.
- Named something "quack.notgit", and it worked fine.
- Cloned Mercurial and Git repositories over SSH with ".git" and non-".git" variants without hitting any issues.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T11902
Differential Revision: https://secure.phabricator.com/D16908
Summary: Fixes T11866. This got converted wrong when doing the `/source/` stuff.
Test Plan: Browsed the root directory of a Subversion repository in Diffusion.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T11866
Differential Revision: https://secure.phabricator.com/D16860
Summary: Fixes T4245. When a repository has a short name, use `/source/shortname/` as its primary URI.
Test Plan:
- Cloned Git repositories from shortnames via HTTP and SSH.
- Cloned Mercurial repositories from shortnames via HTTP and SSH.
- Cloned Subversion repositories from shortnames via SSH.
- Browsed Git, Mercurial and Subversion repositories.
- Added and removed short names to various repositories.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T4245
Differential Revision: https://secure.phabricator.com/D16851
Summary:
This has been replaced by `PolicyCodex` after D16830. Also:
- Rebuild Celerity map to fix grumpy unit test.
- Fix one issue on the policy exception workflow to accommodate the new code.
Test Plan:
- `arc unit --everything`
- Viewed policy explanations.
- Viewed policy errors.
Reviewers: chad
Reviewed By: chad
Subscribers: hach-que, PHID-OPKG-gm6ozazyms6q6i22gyam
Differential Revision: https://secure.phabricator.com/D16831
Summary:
Fixes T11705. I did not realize that `ON DUPLICATE KEY UPDATE` was order-dependent, so the "reset" clause of this `IF(...)` never actually worked.
Reorder it so we check if we're changing the message type //first//, then actually change the message type.
This makes the count reset properly when a failing repository succeeds, or a working repository fails.
Test Plan:
- On `master`, forced a working repository to fail a `bin/repository update`, saw the message change types (expected) but keep the old count (wrong!).
- With this patch, repeated the process and saw the count reset properly.
- Ran the patch, verified counts reset to 0.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T11705
Differential Revision: https://secure.phabricator.com/D16623
Summary:
Fixes T11677. This makes two minor adjustments to the repository import daemons:
- The first step ("Message") now queues at a slightly-lower-than-default (for already-imported repositories) or very-low (for newly importing repositories) priority level.
- The other steps now queue at "default" priority level. This is actually what they already did, but without this change their behavior would be to inherit the priority level of their parents.
This has two effects:
- When adding new repositories to an existing install, they shouldn't block other things from happening anymore.
- The daemons will tend to start one commit and run through all of its steps before starting another commit. This makes progress through the queue more even and predictable.
- Before, they did ALL the message tasks, then ALL the change tasks, etc. This works fine but is confusing/uneven/less-predictable because each type of task takes a different amount of time.
Test Plan:
- Added a new repository.
- Saw all of its "message" steps queue at priority 4000.
- Saw followups queue at priority 2000.
- Saw progress generally "finish what you started" -- go through the queue one commit at a time, instead of one type of task at a time.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T11677
Differential Revision: https://secure.phabricator.com/D16585
Summary:
Ref T11665. Currently, when a repository hits an error, we retry it after 15s. This is correct if the error was temporary/transient/config-related (e.g., bad network or administrator setting up credentials) but not so great if the error is long-lasting (completely bad authentication, invalid URI, etc), as it can pile up to a meaningful amount of unnecessary load over time.
Instead, record how many times in a row we've hit an error and adjust backoff behavior: first error is 15s, then 30s, 45s, etc.
Additionally, when computing the backoff for an empty repository, use the repository creation time as though it was the most recent commit. This is a good proxy which gives us reasonable backoff behavior.
This required removing the `CODE_WORKING` messages, since they would have reset the error count. We could restore them (as a different type of message), but I think they aren't particularly useful since cloning usually doesn't take too long and there's more status information avilable now than there was when this stuff was written.
Test Plan:
- Ran `bin/phd debug pull`.
- Saw sensible, increasing backoffs selected for repositories with errors.
- Saw sensible backoffs selected for empty repositories.
Reviewers: chad
Maniphest Tasks: T11665
Differential Revision: https://secure.phabricator.com/D16575
Summary:
Ref T11665. Fixes T7865. When we restart the daemons, the repository pull daemon currently resets the cooldowns on all of its pulls. This can generate a burst of initial load when restarting a lot of instance daemons (as in the Phacility cluster), described in T7865. This smooths things out so that recent pulls are considered, and any repositories which were waiting keep waiting.
Somewhat counterintuitively, hosted repositories write `TYPE_FETCH` status messages, so this should work equally well for hosted and observed repositories.
This also paves the way for better backoff behavior on repository errors, described in T11665. The error backoff now uses the same logic that the standard backoff does. The next change will make backoff computation consider recent errors.
(This is technically too large for repositories which have encountered one error and have a low commit rate, but I'll fix that in the following change; this is just a checkpoint on the way there.)
Test Plan: Ran `bin/phd debug pull`, saw the daemon compute reasonable windows based on previous pull activity.
Reviewers: chad
Maniphest Tasks: T7865, T11665
Differential Revision: https://secure.phabricator.com/D16574
Summary:
Ref T11650. Currently, we load packages and then discard the archived ones.
However, this gets "dominion" rules (where a more-general package gives up ownership if a more-specific package exists) wrong if the more-specific package is archived: we incorrectly give up ownership.
Instead, just ignore these packages completely when loading affected packages. This is slightly simpler.
(There are technically two pieces of code we have to do this for, which should be a single piece of code but which haven't yet been unified.)
Test Plan:
- Created packages:
- Package A, on "/" (strong dominion, autoreview).
- Package B, on "/x/" (weak dominion, autoreview).
- Package C, on "/x/y" (archived, autoreview).
- Create a revision affecting "/x/y".
- Saw correct path ownership in table of contents ("B", strongest package only).
- Saw correct autoreview behavior (A + B).
- (Prior to patch, in `master`, reproduced the problem behaviors described in T11650, with bad dominion rules and failure to autoreview B.)
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T11650
Differential Revision: https://secure.phabricator.com/D16564
Summary: Ref T11559. This makes managing large numbers of repositories slightly easier.
Test Plan: {F1796119}
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T11559
Differential Revision: https://secure.phabricator.com/D16472
Summary:
Fixes T10423. Ref T11524. This changes `diffusion.rawdiffquery` to return a file PHID instead of a blob of data.
This is better in general, but particularly better for huge diffs (as in T10423) and diffs with non-utf8 data (as in T10423).
Test Plan:
- Used `bin/differential extract` to extract a latin1 diff, got a clean diff.
- Used `bin/repository reparse --herald` to rerun herald on a latin1 diff, got a clean result.
- Pushed latin1 diffs to test commit hooks.
- Triggered the the too large / too slow logic.
- Viewed latin1 diffs in Diffusion.
- Used "blame past this change" in Diffusion to hit the `before` logic.
Reviewers: chad
Reviewed By: chad
Subscribers: eadler
Maniphest Tasks: T10423, T11524
Differential Revision: https://secure.phabricator.com/D16460
Summary:
Fixes T11537. See that task for discussion.
Although we could accommodate these faithfully, it requires a huge migration and affects one repository on one install which was written with buggy tools.
At least for now, just replace out-of-32-bit-range epoch values with the current time, which is often somewhat close to the real value.
Test Plan:
- Following the instructions in T11537, created commits in 40,000 AD.
- Tried to import them, reproducing the "epoch" database issue.
- Applied the patch.
- Successfully imported future-commits, with some liberties around commit dates. Note that author date (not stored in an `epoch` column) is still shown faithfully:
{F1789302}
Reviewers: chad, avivey
Reviewed By: avivey
Maniphest Tasks: T11537
Differential Revision: https://secure.phabricator.com/D16456
Summary:
Ref T11522. This tries to reduce the cost of rewriting a repository by making handles smarter about rewritten commits.
When a handle references an unreachable commit, try to load a rewrite hint for the commit. If we find one, change the handle name to "OldHash > NewHash" to provide a strong hint that the commit was rewritten and that copy/pasting the old hash (say, to the CLI) won't work.
I think this notation isn't totally self-evident, but users can click it to see the big error message on the page, and it's at least obvious that something weird is going on, which I think is the important part.
Some possible future work:
- Not sure this ("Recycling Symbol") is the best symbol? Seems sort of reasonable but mabye there's a better one.
- Putting this information directly on the hovercard could help explain what this means.
Test Plan: {F1780719}
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T11522
Differential Revision: https://secure.phabricator.com/D16437
Summary:
Ref T11522. When a commit is no longer reachable from any branch/tag, we currently show a "this has been deleted" message.
Instead, go further: check if there is a "rewritten" hint pointing at a commit the current commit was rewritten into. If we find one, show a message about that instead.
(This isn't super pretty, just getting it working for now. I expect to revisit this UI in T9713 if we don't get to it before that.)
Test Plan: {F1780703}
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T11522
Differential Revision: https://secure.phabricator.com/D16436
Summary: Ref T11522. This migrates any "badcommit" data (which probably only exists at Facebook and on 1-2 other installs in the wild) to the new "hint" table.
Test Plan:
- Wrote some bad commit annotations to the badcommit table.
- Viewed them in the web UI and used `bin/repository reparse --change ...` to reparse them. Saw "this is bad" messages.
- Ran migration, verified that valid "badcommit" rows were successfully migrated to become "hint" rows.
- Viewed the new web UI and re-parsed the change, saw "unreadable commit" messages.
- Viewed a good commit; reparsed a good commit.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T11522
Differential Revision: https://secure.phabricator.com/D16435
Summary:
Ref T11522. This provides storage for tracking rewritten commits (new feature) and unreadable commits (existing feature, but really hacky).
This doesn't do anything yet, just adds a table and a CLI tool for updating it. I'll document the tool once it works. You just pipe in some JSON, but I need to document the format.
Test Plan:
- Piped JSON for "none", "rewritten" and "unreadable" hints into `bin/repository hint`.
- Examined the database to see that the table was written properly.
- Tried to pipe bad JSON in, invalid hint types, etc. Got reasonable human-readable error messages.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T11522
Differential Revision: https://secure.phabricator.com/D16434
Summary:
Ref T7148. The automated export process runs this via daemon, which can't answer "Y" to this prompt. Let it "--force" instead.
(Some of my test instances didn't have any repositories, which is why I didn't catch this sooner.)
Test Plan: Ran `bin/repository move-paths --force ...`, saw change applied without a prompt.
Reviewers: chad
Reviewed By: chad
Maniphest Tasks: T7148
Differential Revision: https://secure.phabricator.com/D16426
Summary:
Fixes T11453. Currently, commit message summaries are limited to 80 bytes. This may only be 20-40 characters for CJK languages or langauges with Cyrillic script.
Increase storage size to 255, then truncate to the shorter of 255 bytes or 80 glyphs. This preserves the same behavior for latin languages, but is less tight for Russian, etc.
Some minor additional changes:
- Provide a way to ask "how much data fits in this column?" so we don't have to duplicate column lengths across summary checks or UI errors like "title too long".
- Remove the `text80` datatype, since no other columns use it and we have no use cases (or likely use cases) for it.
Test Plan:
- Made a commit with a Cyrillic title, saw reasonable summarization in UI:
{F1757522}
- Added and ran unit tests.
- Grepped for removed `SUMMARY_MAX_LENGTH` constant.
- Grepped for removed `text80` data type.
Reviewers: avivey, chad
Reviewed By: avivey
Subscribers: avivey
Maniphest Tasks: T11453
Differential Revision: https://secure.phabricator.com/D16385
