1 Towards replacing rpm + yum (0.1):
3 - drop the filelists from the main package set file, split out to a
4 secondary file. for package sets that depend on other package sets,
5 we need to be able to generate properties with owning packages that
6 are in another set. this way, a package that requires a file, will
7 look up the provides in the set and find the package that owns it
8 and then try mark that for update.
12 - pre install check; check that dirs can be created (no files where
13 want to create dirs), move config files according to file
16 - store rpm headers for installed packages.
18 - implement rpm uninstall and update.
20 - triggers? just say no?
22 - rpm seems to consider glibc > 2.6.90 to mean greater than
23 2.6.90-anything. That is, a comparison that doesn't mention the
24 release field, shouldn't regard the release field of pkgs it
25 compares against. glibc-common-2.6.90 has
27 conflicts: glibc < 2.6.90, glibc > 2.6.90
29 since rpm doesn't let you do glibc != 2.6.90, and
31 requires: glibc = 2.6.90
33 will always pull in glibc. But even with a != relation, would
34 glibc-2.6.90-16 be equal to 2.6.90? glibc 2.7.90-8 dropped it in
35 favor of requires = 2.7.90-8 (#225806).
39 - space calculation before transaction, but ideally, do a number of
42 - pre-link changing binaries and libs on disk screwing up checksum?
44 - pipelined download and install; topo-sort packages in update set,
45 pick one with all deps in the current set, add it to the current set
46 and satisfy deps against update set => result: minimal update
47 transaction. Queue download and install/update transaction for the
48 packages in the minimal set, start over. This also makes the
49 installation phase much more interruptible, basically just stop
50 after a sub-transaction finishes. As we keep the update set around
51 as a target, we can restart if needed. Probably don't need to, can
52 just do a new update. During a sub-transaction we should keep the
53 target set (i.e. the current set to be) around as a lock file
54 (system.rzdb.lock or so, see git) so that razor updates are
55 prevented if the systems crashes during an update.
57 - implement depsolving between multiple package sets by creating an
58 iterator that has a sorted list of all installed pkgs from all sets,
59 all installed requires from all sets, all installed provides from
60 all sets etc. could be a list of tuples (pkgs index, set index).
61 should simplify even the two-set depsolving a bit since we can
62 pretend there's just one set. this should also be useful for the
63 'overlay set' idea where the system set is actually made up of a
64 number of sets, but typically a read-only set from a read-only fs
65 and a read-write set from a r/w fs.
67 - locking: we use advisory file locking on the system set
68 (/var/lib/razor/system.rzdb) to indicate a transaction is in
69 progress. The locking algorithm is as follows:
71 1. obtain advisory lock on system set. if this is already taken,
72 we know that a process is actively modifying the system set and
73 we have to wait. there's a fcntl that lets you block for the
76 2. if a system-next.rzdb file already exists an earlier razor
77 process was interrupted or crashed and we may want to clean
78 that up. the system-next.rzdb file will record what the
79 previous instance was trying to do and we can just replay that
82 3. create the new package set whichever way and write it to
83 system-next.rzdb, then start installing/removing rpms.
85 4. When the update is complete, rename system-next.rzdb to
86 system.rzdb and remove the advisory lock.
88 we should probably introduce a new object that encapsulates this
89 sequence, the filename conventions, rpm cache, e.g. struct
90 razor_image, with operations such as
92 #define RAZOR_IMAGE_READ 0x01
93 #define RAZOR_IMAGE_WRITE 0x02
96 razor_image_open(const char *root, unsigned int flags);
99 razor_image_begin_transaction(struct razor_image *image,
100 struct razor_set *target);
103 razor_image_finish_transaction(struct razor_image *image);
105 the transaction pipelineing described above sits on top of this,
106 since each step there needs to complete a full transaction that
107 writes out a new package set.
109 for overlay package sets we could do something like
112 razor_image_open_with_base(const char *root, unsigned int flags,
113 struct razor_image *base);
115 where base specifies the r/o package set it's layered on. this
116 allows for stacking several layers of images.
119 Package set file format items:
121 - nail down byte-order of repo file.
123 - version the sections in the file, put the element size in the header
124 so we can add stuff to elements in a backwards compatible way.
125 maybe not necessary, we can just add sections that augment the
126 sections we want to add to (similar to how rpm has add versioned
132 - keep history of installed packages/journal of package transaction,
133 so we can roll back to yesterday, or see what got installed in the
136 - use hash table for package and property lists so we only store
137 unique lists (like for string pool).
139 - use existing, running system as repo; eg
141 razor update razor://other-box.local evince
143 to pull eg the latest evince and dependencies from another box. We
144 should be able to regenerate a rzr pkg from the system so we can
145 reuse the signature from the originating repo.
147 - Ok, maybe the fastest package set merge method in the end is to use
148 the razor_importer, but use a hash table for the properties. This
149 way we can assign them unique IDs immediately (like tokenizing
152 - test suite should be easy, just keep .rzdb files around and test
153 different type of upgrades that way (obsoletes, conflicts, file
154 conflicts, file/dir problems etc). Or maybe just keep a simple file
155 format ad use a custom importer to create the .rzdb files.
157 - overlay package sets? mount a read-only /usr over nfs or from the
158 virt-host and have a local package set overlaid over the read-only
159 one. shouldn't need new features in the core package set data
160 structure, but should be just conventions on top. we have the base
161 package set from the r/o system, the overlay set from the local
162 system and we can have an effective package set which is the merge
163 of everything from the overlay into the base set. the effective set
164 is easy to compute and we could do it on the fly or cache it. or
165 maybe the effective set is the on-disk representation and the
166 overlay can be computed when needed, we just keep a link back to the
169 - incremental rawhide repo updates? instead of downloading 10MB zipped
170 repo every time, download a diff rzdb? Should be pretty small,
171 especially if we don't have file checksums in metadata. Filenames
172 and properties are for the most part already present, typically just
173 a version bump plus maybe tweaking a couple requires. The upstream
174 repo can store multiple incremental updates in one big file and
175 provide an index file that maps updates for a given date (we should
176 use repo-file checksums though) to a range in the file: Download the
177 index file, search for a match for your latest rawhide.rzdb file,
178 download range of updates that brings it up to date.
180 - use hash tables for dirs when importing files to avoid qsorting all
185 - eliminate duplicate entries in package property lists.