netcdf-c/libnczarr/zmap_s3sdk.c
Dennis Heimbigner df3636b959 Mitigate S3 test interference + Unlimited Dimensions in NCZarr
This PR started as an attempt to add unlimited dimensions to NCZarr.
It did that, but this exposed significant problems with test interference.
So this PR is mostly about fixing -- well mitigating anyway -- test
interference.

The problem of test interference is now documented in the document docs/internal.md.
The solutions implemented here are also describe in that document.
The solution is somewhat fragile but multiple cleanup mechanisms
are provided. Note that this feature requires that the
AWS command line utility must be installed.

## Unlimited Dimensions.
The existing NCZarr extensions to Zarr are modified to support unlimited dimensions.
NCzarr extends the Zarr meta-data for the ".zgroup" object to include netcdf-4 model extensions. This information is stored in ".zgroup" as dictionary named "_nczarr_group".
Inside "_nczarr_group", there is a key named "dims" that stores information about netcdf-4 named dimensions. The value of "dims" is a dictionary whose keys are the named dimensions. The value associated with each dimension name has one of two forms
Form 1 is a special case of form 2, and is kept for backward compatibility. Whenever a new file is written, it uses format 1 if possible, otherwise format 2.
* Form 1: An integer representing the size of the dimension, which is used for simple named dimensions.
* Form 2: A dictionary with the following keys and values"
   - "size" with an integer value representing the (current) size of the dimension.
   - "unlimited" with a value of either "1" or "0" to indicate if this dimension is an unlimited dimension.

For Unlimited dimensions, the size is initially zero, and as variables extend the length of that dimension, the size value for the dimension increases.
That dimension size is shared by all arrays referencing that dimension, so if one array extends an unlimited dimension, it is implicitly extended for all other arrays that reference that dimension.
This is the standard semantics for unlimited dimensions.

Adding unlimited dimensions required a number of other changes to the NCZarr code-base. These included the following.
* Did a partial refactor of the slice handling code in zwalk.c to clean it up.
* Added a number of tests for unlimited dimensions derived from the same test in nc_test4.
* Added several NCZarr specific unlimited tests; more are needed.
* Add test of endianness.

## Misc. Other Changes
* Modify libdispatch/ncs3sdk_aws.cpp to optionally support use of the
   AWS Transfer Utility mechanism. This is controlled by the
   ```#define TRANSFER```` command in that file. It defaults to being disabled.
* Parameterize both the standard Unidata S3 bucket (S3TESTBUCKET) and the netcdf-c test data prefix (S3TESTSUBTREE).
* Fixed an obscure memory leak in ncdump.
* Removed some obsolete unit testing code and test cases.
* Uncovered a bug in the netcdf-c handling of big-endian floats and doubles. Have not fixed yet. See tst_h5_endians.c.
* Renamed some nczarr_tests testcases to avoid name conflicts with nc_test4.
* Modify the semantics of zmap\#ncsmap_write to only allow total rewrite of objects.
* Modify the semantics of zodom to properly handle stride > 1.
* Add a truncate operation to the libnczarr zmap code.
2023-09-26 16:56:48 -06:00

590 lines
15 KiB
C

/*
* Copyright 2018, University Corporation for Atmospheric Research
* See netcdf/COPYRIGHT file for copying and redistribution conditions.
*/
#include "zincludes.h"
#include "zmap.h"
#include "ncs3sdk.h"
#undef S3DEBUG
/*
Map our simplified map model to an S3 bucket + objects.
For the API, the mapping is as follows:
1. A bucket contains multiple datasets.
2. A key to the root of a dataset is assumed to be with respect to
the bucket name from the original open/create url.
2. Containment is simulated using the S3 key conventions.
3. Every object (e.g. group or array) is mapped to an S3 object
4. Meta data objects (e.g. .zgroup, .zarray, etc) are kept as an S3 object.
5. Actual variable data (for e.g. chunks) is stored as
using an S3 object per chunk.
Given a key, one of three things is true.
1. The key points to a content-bearing object
2. The key has no associated object
This basically means that there is no notion of not-found:
all keys are assumed to exist, but may have no content.
Notes:
1. Our canonical URLs use path style rather than virtual-host, although
virtual-host URLs will be accepted anc converted to path-style.
*/
#undef DEBUG
#define DEBUGERRORS
#define NCZM_S3SDK_V1 1
#define ZS3_PROPERTIES (0)
/* Define the "subclass" of NCZMAP */
typedef struct ZS3MAP {
NCZMAP map;
NCS3INFO s3;
void* s3client;
char* errmsg;
} ZS3MAP;
/* Forward */
static NCZMAP_API nczs3sdkapi; // c++ will not allow static forward variables
static int zs3len(NCZMAP* map, const char* key, size64_t* lenp);
static void freevector(size_t nkeys, char** list);
static void zs3initialize(void);
static int s3clear(void* s3client, const char* bucket, const char* key);
static int maketruekey(const char* rootpath, const char* key, char** truekeyp);
static void
errclear(ZS3MAP* z3map)
{
if(z3map) {
if(z3map->errmsg)
free(z3map->errmsg);
z3map->errmsg = NULL;
}
}
#ifdef DEBUGERRORS
static void
reporterr(ZS3MAP* z3map)
{
if(z3map) {
if(z3map->errmsg) {
nclog(NCLOGERR,z3map->errmsg);
}
errclear(z3map);
}
}
#else
#define reporterr(map)
#endif
/* Define the Dataset level API */
static int zs3initialized = 0;
static void
zs3initialize(void)
{
if(!zs3initialized) {
ZTRACE(7,NULL);
zs3initialized = 1;
(void)ZUNTRACE(NC_NOERR);
}
}
void
NCZ_s3finalize(void)
{
zs3initialized = 0;
}
static int
zs3create(const char *path, int mode, size64_t flags, void* parameters, NCZMAP** mapp)
{
int stat = NC_NOERR;
ZS3MAP* z3map = NULL;
NCURI* url = NULL;
char* prefix = NULL;
char* truekey = NULL;
NC_UNUSED(flags);
NC_UNUSED(parameters);
ZTRACE(6,"path=%s mode=%d flag=%llu",path,mode,flags);
if(!zs3initialized) zs3initialize();
/* Build the z3 state */
if((z3map = (ZS3MAP*)calloc(1,sizeof(ZS3MAP))) == NULL)
{stat = NC_ENOMEM; goto done;}
z3map->map.format = NCZM_S3;
z3map->map.url = strdup(path);
z3map->map.mode = mode;
z3map->map.flags = flags;
z3map->map.api = (NCZMAP_API*)&nczs3sdkapi;
/* Parse the URL */
ncuriparse(path,&url);
if(url == NULL)
{stat = NC_EURL; goto done;}
/* Convert to canonical path-style */
if((stat = NC_s3urlprocess(url,&z3map->s3,NULL))) goto done;
/* Verify the root path */
if(z3map->s3.rootkey == NULL)
{stat = NC_EURL; goto done;}
z3map->s3client = NC_s3sdkcreateclient(&z3map->s3);
{
int exists = 0;
/* Does bucket already exist */
if((stat = NC_s3sdkbucketexists(z3map->s3client,z3map->s3.bucket,&exists, &z3map->errmsg))) goto done;
if(!exists) {
/* create it */
if((stat = NC_s3sdkbucketcreate(z3map->s3client,z3map->s3.region,z3map->s3.bucket,&z3map->errmsg)))
goto done;
}
/* The root object may or may not already exist */
switch (stat = NC_s3sdkinfo(z3map->s3client,z3map->s3.bucket,z3map->s3.rootkey,NULL,&z3map->errmsg)) {
case NC_EEMPTY: /* no such object */
stat = NC_NOERR; /* which is what we want */
errclear(z3map);
break;
case NC_NOERR:
stat = NC_EOBJECT;
errclear(z3map);
/* Delete objects inside root object tree */
s3clear(z3map->s3client,z3map->s3.bucket,z3map->s3.rootkey);
goto done; /* already exists */
default: reporterr(z3map); goto done;
}
}
if(mapp) *mapp = (NCZMAP*)z3map;
done:
reporterr(z3map);
ncurifree(url);
nullfree(prefix);
nullfree(truekey);
if(stat) nczmap_close((NCZMAP*)z3map,1);
return ZUNTRACE(stat);
}
/* The problem with open is that there
no obvious way to test for existence.
So, we assume that the dataset must have
some content. We look for that */
static int
zs3open(const char *path, int mode, size64_t flags, void* parameters, NCZMAP** mapp)
{
int stat = NC_NOERR;
ZS3MAP* z3map = NULL;
NCURI* url = NULL;
NClist* content = NULL;
size_t nkeys = 0;
NC_UNUSED(flags);
NC_UNUSED(parameters);
ZTRACE(6,"path=%s mode=%d flags=%llu",path,mode,flags);
if(!zs3initialized) zs3initialize();
/* Build the z3 state */
if((z3map = (ZS3MAP*)calloc(1,sizeof(ZS3MAP))) == NULL)
{stat = NC_ENOMEM; goto done;}
z3map->map.format = NCZM_S3;
z3map->map.url = strdup(path);
z3map->map.mode = mode;
z3map->map.flags = flags;
z3map->map.api = (NCZMAP_API*)&nczs3sdkapi;
/* Parse the URL */
if((stat = ncuriparse(path,&url))) goto done;
if(url == NULL)
{stat = NC_EURL; goto done;}
/* Convert to canonical path-style */
if((stat = NC_s3urlprocess(url,&z3map->s3,NULL))) goto done;
/* Verify root path */
if(z3map->s3.rootkey == NULL)
{stat = NC_EURL; goto done;}
z3map->s3client = NC_s3sdkcreateclient(&z3map->s3);
/* Search the root for content */
content = nclistnew();
if((stat = NC_s3sdkgetkeys(z3map->s3client,z3map->s3.bucket,z3map->s3.rootkey,&nkeys,NULL,&z3map->errmsg)))
goto done;
if(nkeys == 0) {
/* dataset does not actually exist; we choose to return ENOOBJECT instead of EEMPTY */
stat = NC_ENOOBJECT;
goto done;
}
if(mapp) *mapp = (NCZMAP*)z3map;
done:
reporterr(z3map);
nclistfreeall(content);
ncurifree(url);
if(stat) nczmap_close((NCZMAP*)z3map,0);
return ZUNTRACE(stat);
}
/* This uses url so we can get bucket */
static int
zs3truncate(const char *s3url)
{
int stat = NC_NOERR;
void* s3client = NULL;
NCURI* url = NULL;
NCURI* purl = NULL;
NCS3INFO info;
ZTRACE(6,"url=%s",s3url);
ncuriparse(s3url,&url);
if(url == NULL) {stat = NC_EURL; goto done;}
if((stat=NC_s3urlprocess(url,&info,&purl))) goto done;
if((s3client = NC_s3sdkcreateclient(&info))) {stat = NC_ES3; goto done;}
if((stat = s3clear(s3client,info.bucket,purl->path))) goto done;
done:
if(s3client) {stat=NC_s3sdkclose(s3client,&info,1,NULL);}
ncurifree(url);
ncurifree(purl);
(void)NC_s3clear(&info);
return stat;
}
/**************************************************/
/* Object API */
/*
@return NC_NOERR if key points to a content-bearing object.
@return NC_EEMPTY if object at key has no content.
@return NC_EXXX return true error
*/
static int
zs3exists(NCZMAP* map, const char* key)
{
int stat = NC_NOERR;
ZTRACE(6,"map=%s key=%s",map->url,key);
stat = zs3len(map,key,NULL);
return ZUNTRACE(stat);
}
/*
@return NC_NOERR if key points to a content-bearing object.
@return NC_EEMPTY if object at key has no content.
@return NC_EXXX return true error
*/
static int
zs3len(NCZMAP* map, const char* key, size64_t* lenp)
{
int stat = NC_NOERR;
ZS3MAP* z3map = (ZS3MAP*)map;
char* truekey = NULL;
ZTRACE(6,"map=%s key=%s",map->url,key);
if((stat = maketruekey(z3map->s3.rootkey,key,&truekey))) goto done;
switch (stat = NC_s3sdkinfo(z3map->s3client,z3map->s3.bucket,truekey,lenp,&z3map->errmsg)) {
case NC_NOERR: break;
case NC_EEMPTY:
if(lenp) *lenp = 0;
goto done;
default:
goto done;
}
done:
nullfree(truekey);
reporterr(z3map);
return ZUNTRACE(stat);
}
/*
@return NC_NOERR if object at key was read
@return NC_EEMPTY if object at key has no content.
@return NC_EXXX return true error
*/
static int
zs3read(NCZMAP* map, const char* key, size64_t start, size64_t count, void* content)
{
int stat = NC_NOERR;
ZS3MAP* z3map = (ZS3MAP*)map; /* cast to true type */
size64_t size = 0;
char* truekey = NULL;
ZTRACE(6,"map=%s key=%s start=%llu count=%llu",map->url,key,start,count);
if((stat = maketruekey(z3map->s3.rootkey,key,&truekey))) goto done;
switch (stat=NC_s3sdkinfo(z3map->s3client, z3map->s3.bucket, truekey, &size, &z3map->errmsg)) {
case NC_NOERR: break;
case NC_EEMPTY: goto done;
default: goto done;
}
/* Sanity checks */
if(start >= size || start+count > size)
{stat = NC_EEDGE; goto done;}
if(count > 0) {
if((stat = NC_s3sdkread(z3map->s3client, z3map->s3.bucket, truekey, start, count, content, &z3map->errmsg)))
goto done;
}
done:
nullfree(truekey);
reporterr(z3map);
return ZUNTRACE(stat);
}
/*
@return NC_NOERR if key content was written
@return NC_EEMPTY if object at key has no content.
@return NC_EXXX return true error
*/
static int
zs3write(NCZMAP* map, const char* key, size64_t count, const void* content)
{
int stat = NC_NOERR;
ZS3MAP* z3map = (ZS3MAP*)map; /* cast to true type */
char* chunk = NULL; /* use char* so we can do arithmetic with it */
char* truekey = NULL;
size64_t objsize;
ZTRACE(6,"map=%s key=%s count=%llu",map->url,key,count);
if((stat = maketruekey(z3map->s3.rootkey,key,&truekey))) goto done;
/* Apparently S3 has no write byterange operation, so we need to read the whole object,
copy data, and then rewrite */
switch (stat=NC_s3sdkinfo(z3map->s3client, z3map->s3.bucket, truekey, &objsize, &z3map->errmsg)) {
case NC_NOERR: /* Figure out the new size of the object */
break;
case NC_EEMPTY:
stat = NC_NOERR; /* reset */
break;
default: reporterr(z3map); goto done;
}
chunk = (char*)calloc(1,count); /* initialize it */
if(chunk == NULL)
{stat = NC_ENOMEM; goto done;}
/* overwrite the relevant part of the memory with the contents */
if(count > 0)
memcpy((char*)chunk,content,count);
/* (re-)write */
if((stat = NC_s3sdkwriteobject(z3map->s3client, z3map->s3.bucket, truekey, count, (void*)chunk, &z3map->errmsg)))
goto done;
done:
nullfree(truekey);
reporterr(z3map);
nullfree(chunk);
return ZUNTRACE(stat);
}
static int
zs3close(NCZMAP* map, int deleteit)
{
int stat = NC_NOERR;
ZS3MAP* z3map = (ZS3MAP*)map;
ZTRACE(6,"map=%s deleteit=%d",map->url, deleteit);
if(deleteit)
s3clear(z3map->s3client,z3map->s3.bucket,z3map->s3.rootkey);
if(z3map->s3client && z3map->s3.bucket && z3map->s3.rootkey) {
NC_s3sdkclose(z3map->s3client, &z3map->s3, deleteit, &z3map->errmsg);
}
reporterr(z3map);
z3map->s3client = NULL;
NC_s3clear(&z3map->s3);
nullfree(z3map->errmsg);
nczm_clear(map);
nullfree(map);
return ZUNTRACE(stat);
}
/*
Return a list of full keys immediately "below" a specified prefix,
but not including the prefix.
In theory, the returned list should be sorted in lexical order,
but it possible that it is not.
@return NC_NOERR if success, even if no keys returned.
@return NC_EXXX return true error
*/
static int
zs3search(NCZMAP* map, const char* prefix, NClist* matches)
{
int i,stat = NC_NOERR;
ZS3MAP* z3map = (ZS3MAP*)map;
char** list = NULL;
size_t nkeys;
NClist* tmp = NULL;
char* trueprefix = NULL;
char* newkey = NULL;
const char* p;
ZTRACE(6,"map=%s prefix0=%s",map->url,prefix);
if((stat = maketruekey(z3map->s3.rootkey,prefix,&trueprefix))) goto done;
if(*trueprefix != '/') return NC_EINTERNAL;
if((stat = NC_s3sdkgetkeys(z3map->s3client,z3map->s3.bucket,trueprefix,&nkeys,&list,&z3map->errmsg)))
goto done;
if(nkeys > 0) {
size_t tplen = strlen(trueprefix);
tmp = nclistnew();
/* Remove the trueprefix from the front of all the returned keys */
for(i=0;i<nkeys;i++) {
const char* l = list[i];
if(memcmp(trueprefix,l,tplen)==0) {
p = l+tplen; /* Point to start of suffix */
/* If the key is same as trueprefix, ignore it */
if(*p == '\0') continue;
if(nczm_segment1(p,&newkey)) goto done;
nclistpush(tmp,newkey); newkey = NULL;
}
}
/* Now remove duplicates */
for(i=0;i<nclistlength(tmp);i++) {
int j;
int duplicate = 0;
const char* is = nclistget(tmp,i);
for(j=0;j<nclistlength(matches);j++) {
const char* js = nclistget(matches,j);
if(strcmp(js,is)==0) {duplicate = 1; break;} /* duplicate */
}
if(!duplicate)
nclistpush(matches,strdup(is));
}
nclistfreeall(tmp); tmp = NULL;
}
#ifdef DEBUG
for(i=0;i<nclistlength(matches);i++) {
const char* is = nclistget(matches,i);
fprintf(stderr,"search: %s\n",is);
}
#endif
done:
nullfree(newkey);
nullfree(trueprefix);
reporterr(z3map);
nclistfreeall(tmp);
freevector(nkeys,list);
return ZUNTRACEX(stat,"|matches|=%d",(int)nclistlength(matches));
}
/**************************************************/
/* S3 Utilities */
/*
Remove all objects in bucket with keys which have
rootkey as prefix; rootkey is a truekey
*/
static int
s3clear(void* s3client, const char* bucket, const char* rootkey)
{
int stat = NC_NOERR;
char** list = NULL;
char** p;
size_t nkeys = 0;
if(s3client && bucket && rootkey) {
if((stat = NC_s3sdksearch(s3client, bucket, rootkey, &nkeys, &list, NULL)))
goto done;
if(list != NULL) {
for(p=list;*p;p++) {
/* If the key is the rootkey, skip it */
if(strcmp(rootkey,*p)==0) continue;
#ifdef S3DEBUG
fprintf(stderr,"s3clear: %s\n",*p);
#endif
if((stat = NC_s3sdkdeletekey(s3client, bucket, *p, NULL)))
goto done;
}
}
}
done:
NCZ_freestringvec(nkeys,list);
return THROW(stat);
}
/* Prefix key with path to root to make true key */
static int
maketruekey(const char* rootpath, const char* key, char** truekeyp)
{
int stat = NC_NOERR;
char* truekey = NULL;
size_t len, rootlen, keylen;
if(truekeyp == NULL) goto done;
rootlen = strlen(rootpath);
keylen = strlen(key);
len = (rootlen+keylen+1+1+1);
truekey = (char*)malloc(len+1);
if(truekey == NULL) {stat = NC_ENOMEM; goto done;}
truekey[0] = '\0';
if(rootpath[0] != '/')
strlcat(truekey,"/",len+1);
strlcat(truekey,rootpath,len+1);
if(rootpath[rootlen-1] != '/')
strlcat(truekey,"/",len+1);
if(key[0] == '/') key++;
strlcat(truekey,key,len+1);
if(key[keylen-1] == '/') /* remove any trailing '/' */
truekey[strlen(truekey)-1] = '\0';
*truekeyp = truekey; truekey = NULL;
done:
nullfree(truekey);
return stat;
}
static void
freevector(size_t nkeys, char** list)
{
size_t i;
if(list) {
for(i=0;i<nkeys;i++) nullfree(list[i]);
nullfree(list);
}
}
/**************************************************/
/* External API objects */
NCZMAP_DS_API zmap_s3sdk;
NCZMAP_DS_API zmap_s3sdk = {
NCZM_S3SDK_V1,
ZS3_PROPERTIES,
zs3create,
zs3open,
zs3truncate,
};
static NCZMAP_API
nczs3sdkapi = {
NCZM_S3SDK_V1,
zs3close,
zs3exists,
zs3len,
zs3read,
zs3write,
zs3search,
};